Search

This paper contributes to the ongoing debate on the nature and characteristics of the volatility transmission channels of major crash events in international stock markets between 03 July 1997 and 09 March 2021. Using dynamic conditional correlations (DCC) for conditional correlations and volatility clustering, GARCH-BEKK for the direction of transmission of disturbances, and the Diebold-Yilmaz spillover index for the level of volatility contagion, the paper finds that the climbs in external shock transmissions have long-lasting impacts in domestic markets due to the contagion effect during crisis periods. The findings also reveal that the heavier magnitude of financial stress is transmitted between Asian countries via the Hong Kong stock market. Additionally, the degree of volatility spillovers between advanced and emerging equity markets is smaller compared to the pure spillovers between advanced markets or emerging markets, offering a window of opportunity for international market participants in terms of portfolio diversification and risk management applications. Furthermore, the study introduces a novel early warning system created by integrating DCC correlations with a state-of-the-art deep learning model to predict the global financial crisis and COVID-19 crisis. The experimental analysis of long short-term memory network finds evidence of contagion risk by verifying bursts in volatility spillovers and generating signals with high accuracy before the 12-month crisis period. This provides supplementary information that contributes to the decision-making process of practitioners, as well as offering indicative evidence that facilitates the assessment of market vulnerability for policymakers. [ABSTRACT FROM AUTHOR]

Landslides cause severe economic damage and a large number of casualties every year around the world. In many cases, it is not possible to avoid them, and the task of engineers and geologists is to mitigate their effects using measures such as building diverting structures or preparing escape roads to be used when an alarm is triggered. It is necessary, therefore, to predict the path of the landslide, its depth and velocity, and the runout. These objectives are usually are attained by using mathematical, numerical and rheological models. An important limitation of the analysis is the lack of data, specially when few laboratory tests are available, and in cases where their present important variations. This leads to performing sensitivity analyses in which analysts study the influence of several magnitudes of interest, such as friction angle, porosity, basal pore pressure and geometry of the sliding mass, just to mention a few, leading in turn to perform a large number of simulations. We propose in this paper a methodology to speed up the process, which is based on: (i) using depth integrated models, which provide a good combination of accuracy and computer effort and (ii) using artificial intelligence tools to reduce the number of simulations. Let us consider a case where we have N mag main variables to explore; for each of them we select a number of cases, which can differ from one magnitude to another. The number of cases will be where N cases (i) is the number of cases we have selected for magnitude i. We can consider these variables as nodes belonging to a hypercube of dimension N mag . We will refer from now on as "hypercube" to the set of all cases generated in this way. The paper presents two cases where these techniques will be applied: (i) a 1D dam break problem and (ii) a case of a real debris flow which happened in Hong Kong, for which there is available information. [ABSTRACT FROM AUTHOR]

Different sets of drivers underlie different safety behaviors, and uncovering such complex patterns helps formulate targeted measures to cultivate safety behaviors. Machine learning can explore such complex patterns among safety behavioral data. This paper aims to develop a classification framework for construction personnel's safety behaviors with machine learning algorithms, including logistics regression (LR), support vector machine (SVM), random forest (RF), and categorical boosting (CatBoost). The classification framework has three steps, i.e., data collection and preprocessing, modeling and algorithm implementation, and optimal model acquisition. For illustrative purposes, five common safety behaviors of a random sample of Hong Kong-based construction personnel are used to validate the classification framework. To achieve high classification performance, this paper employed a combinative strategy, consisting of feature selection, synthetic minority over-sampling technique (SMOTE), one-hot encoding, standard scaler and classifiers to classify safety behaviors, and multi-objective slime mould algorithm (MOSMA) to optimize parameters in the classifiers. Results suggest that the combinative strategy of CatBoost–MOSMA achieves the highest classification performance with the maximum average scores, including area under the curve of receiver characteristic operator (AUC) ranging from 0.84 to 0.92, accuracy ranging from 0.80 to 0.86, and F1-score ranging from 0.79 to 0.86. From the optimal model, a unique set of important features was identified for each safety behavior, and ten out of the 46 input indicators were found important for all five safety behaviors. Based on the findings, this study advocates using the machine learning strategy of CatBoost–MOSMA in future construction safety behavior research and makes concrete and targeted suggestions to cultivate different construction safety behaviors. [ABSTRACT FROM AUTHOR]

A major source of risk in green building projects (GBPs) is inaccurate human prediction of the final project cost and duration, which in turn results in cost and schedule overruns (i.e., poor project performance). This paper presents promising new models to mitigate such risk based upon machine learning (ML). Historical data from 198 GBPs in Hong Kong were used to develop and train two fully connected deep neural networks (DNN) models to learn and predict cost and duration, respectively, based on green building rating (GBR) and other project parameters. The models can predict cost and duration with mean absolute percentage error (MAPE) values of 0.07 and 0.09, respectively. They were then integrated with support vector regression (SVR), and results indicated that the integrated DNN-SVR models improve prediction accuracy, decreasing the MAPE from 0.07 to 0.06 (cost) and 0.09 to 0.07 (duration), respectively. The validated models were for the first time deployed as a ML-based web application for automated, fast, and accurate GBP cost and duration prediction. The feature importance analysis results revealed that the most influential parameters on the GBP cost and duration are project area and weather, respectively, not the GBR. Theoretically, the outcomes of this study provide new insights into the impact of GBR on project cost and duration, which are useful for the promotion of GBPs to improve sustainability. Practically, the study provides policymakers and practitioners with novel ML-based models and a web application to improve GBP delivery performance. Green building projects help to combat climate change and improve our health, wellbeing, and quality of life, but they face two key challenges in their execution: cost overruns and schedule delays. To address these challenges, there is a need for accurate prediction of the final project cost and duration from the early stages of the project. The practical relevance of this study is in the development of the first data-driven and machine learning-based web application for addressing this need. New integrated optimized machine learning models are developed. For practitioners to have access to and use these models without the need to possess machine learning expertise, a corresponding easy-to-use web application is offered. From the design and construction stages of their project, practitioners only need to input the green building ratings in sustainable site, materials and waste, energy use, water use, health and wellbeing, and innovations and additions they want to achieve for the project. The project type and area (type and size), original budget, planned duration, and start month (SM) and year should also be input. Once this project data input is completed, the web application automatically and instantaneously predicts with a high level of accuracy the total costs and time needed to deliver the project successfully—on time and on budget. These cost and duration prediction outputs are a valuable tool in helping practitioners adjust green building project plans and budgets to develop more realistic and accurate project budgets and programs to avoid cost overruns and schedule delays. [ABSTRACT FROM AUTHOR]

Accurate storm surge forecasting is vital for saving lives and avoiding economic and infrastructural damage. Failure to accurately predict storm surge can have catastrophic repercussions. Advances in machine learning models show the ability to improve accuracy of storm surge prediction by leveraging vast amounts of historical and realtime data such as weather and tide patterns. This paper proposes a bidirectional attention-based LSTM storm surge architecture (BALSSA) to improve prediction accuracy. Training and evaluation utilized extensive meteorological and tide level data from 77 typhoon incidents in Hong Kong and Macao between 2017 and 2022. The proposed methodology is able to model complex non-linearities between large amounts of data from different sources and identify complex relationships between variables that are typically not captured by traditional physical methods. BALSSA effectively resolves the problem of long-term dependencies in storm surge prediction by the incorporation of an attention mechanism. It enables selective emphasis on significant features and boosts the prediction accuracy. Evaluation has been conducted using real-world datasets from Macao to validate our storm surge prediction model. Results show that accuracy and robustness of predictions were significantly improved by the incorporation of attention mechanisms in our models. BALSSA captures temporal dynamics effectively, providing highly accurate storm surge forecasts (MAE: 0.0126, RMSE: 0.0003) up to 72 h in advance. These findings have practical significance for disaster risk reduction strategies, saving lives through timely evacuation and early warnings. Experiments comparing BALSSA variations with other machine learning algorithms consistently validate BALSSA's superior predictive performance. It offers an additional risk management tool for civil-protection agencies and governments, as well as an ideal solution for enhancing storm surge prediction accuracy, benefiting coastal communities. [ABSTRACT FROM AUTHOR]

Hong Kong International Airport (HKIA) is an important sea airdrome in China. The aim of this study is to evaluate the impacts of this reclamation on the water quality of the Northwestern Bay of Hong Kong (NWBHK). In all, 117 Landsat 5 TM and 44 Landsat 8 OLI images were preprocessed and matched with the marine water data of 18 in situ monitoring points, acquiring 458 and 119 sets of data, respectively. This study adopted BPNN Machine Learning methods to establish the retrieval algorithm. Based on the images, the construction of HKIA was divided into three stages: (1) the construction of the first and second runways from 1992 to 1995; (2) the construction of the Hong Kong Port from 2013 to 2016; and (3) the construction of the third runway from 2017 to 2020. The concentrations of suspended particulate matter, orthophosphate phosphorus, and dissolved inorganic nitrogen from 1986 to 2022 were retrieved. In this paper, it was found that (1) the construction activities led to an increase in SPM, PO4P, and DIN concentrations in adjacent water bodies; (2) the impact of the Tuen Mun River on the NWBHK increased, while the impact of the Tung Chung River on the NWBHK decreased; and (3) the interception impact of HKIA on the transportation of the Pearl River water became stronger. [ABSTRACT FROM AUTHOR]

This paper shows that investments based on deep learning signals extract profitability from difficult-to-arbitrage stocks and during high limits-to-arbitrage market states. In particular, excluding microcaps, distressed stocks, or episodes of high market volatility considerably attenuates profitability. Machine learning-based performance further deteriorates in the presence of reasonable trading costs because of high turnover and extreme positions in the tangency portfolio implied by the pricing kernel. Despite their opaque nature, machine learning methods successfully identify mispriced stocks consistent with most anomalies. Beyond economic restrictions, deep learning signals are profitable in long positions and recent years and command low downside risk. This paper was accepted by Kay Giesecke, finance. Funding: D. Avramov acknowledges the Israel Science Foundation (Grant 288/18) for financial support. S. Cheng acknowledges the General Research Fund of the Research Grants Council of Hong Kong [Project 14502318] for financial support. Supplemental Material: The data files and online appendix are available at https://doi.org/10.1287/mnsc.2022.4449. [ABSTRACT FROM AUTHOR]

Rock classification provides vital information to geosciences and geological engineering practices. Reaping the benefits of the advances of computer vision-based deep learning artificial intelligence (AI) technology, this study aims to develop a next-generation convolutional neural network (CNN) to perform automatic rock classification. Two major challenging issues have been particularly addressed. First, most of the previous rock classifications are simply transfer learning of CNNs that are trained by life-like scenarios. Second, classifying rock types with similar textures leads to severe overfitting of CNNs. In this study, a novel CNN called HKUDES_Net is proposed and implemented to classify seven common Hong Kong rock types, namely fine-grained granite, medium-grained granite, coarse-grained granite, coarse ash tuff, fine ash tuff, feldsparphyric rhyolite, and granodiorite. With the aid of dynamic expansion, squeeze and excitation, and other strategies, HKUDES_Net can classify rock types with similar texture patterns/colors but different grain sizes. As compared with the other ten landmark CNNs and seven feature-based algorithms, HKUDES_Net has the best performance in precision (90.9%), recall (90.4%), and f1-score (90.5%). By implementing the alerting level, which restricts the training loss hovering above a small constant and prevents the validation loss from rising, the overfitting has been efficiently eliminated. The proposal and implementation of HKUDES_Net highlight the value of interdisciplinary research and will continuously pave the way for better coupling AI and geosciences. Highlights: A next-generation convolutional neural network (CNN) called HKUDES_Net is developed to perform automatic rock classification. HKUDES_Net can classify rock types with similar texture patterns/colors but different grain sizes. By implementing the alerting level in HKUDES_Net, the overfitting has been efficiently eliminated. Plain Language Summary: Rock classification provides vital information to geosciences and geological engineering practices. However, the traditional way of rock classification is to a certain degree experience-based. Computer vision-based artificial intelligence (AI), especially convolutional neural network (CNN), a sub-branch of AI, may help people automatically classify rock types with the hope to train AI models (CNNs), which can instantly and precisely perceive rock images similar to how human beings observe rocks. In other words, by putting a rock image into a CNN, the CNN will read the information from the image and tell people what rock type in the image is. This study develops and implements a next-generation CNN called HKUDES_Net to classify seven different rock types. Adopting different computational strategies, HKUDES_Net can classify rock types of different grain sizes with similar texture patterns/colors, which is challenging for other CNNs. As compared with the other landmark CNNs and machine learning algorithms, HKUDES_Net has much a better prediction accuracy performance. The proposal and implementation of HKUDES_Net exemplify an interdisciplinary research study, and we are confident that this paper will be instrumental to pave the way for further coupling AI and geosciences. [ABSTRACT FROM AUTHOR]

With the opening of the Stock Connect programs, the mainland China and Hong Kong stock markets are becoming more closely linked. In this paper, we develop a China's stock market risk early warning system. The proposed early warning system consists of three components. First, we use value at risk (VaR) to identify the stock market risk in which stock market risk is divided into multiple categories instead of two categories. Second, we construct a comprehensive indicator system in which basic indicators, technical indicators, overseas return rate indicators, and macroeconomic indicators are considered simultaneously. Third, we use four machine learning models, namely long short-term memory (LSTM), gate recurrent unit (GRU), multilayer perceptron (MLP), and EXtreme Gradient Boosting algorithm (XGBoost), to predict China's stock market risk. Experimental results show that: (1) Considering the macroeconomic indicators and basic indicators of Shanghai Composite Index (SSEC), ShenZhen Component Index (SZCZ) and Hang Seng Index (HSI) can significantly improve the performance of predicting China's stock market risk. (2) The opening of SH-HK Stock Connect program improves the predictive performance, but the opening of SZ-HK Stock Connect program decreases the predictive performance. (3) The indicators related to Hong Kong become more important after the SZ-HK Stock Connect program. [ABSTRACT FROM AUTHOR]

Intelligently predicting the frequency and volume of natural hazards, including boulder falls, attracts widespread attention in earth science and engineering communities. Taking Hong Kong, China, which is a highly developed and densely populated city, as a research example, the present paper statistically studies the occurrence of boulder falls based on power law relationships and supervised machine learning. The 2002–2016 boulder fall inventory of Hong Kong, which consists of 194 boulder falls, is compiled and analyzed in detail, including volume and spatial distributions. Based on the findings, a power law relationship with the b-value greater than one is particularly proposed for Hong Kong, which then allows major boulder falls (volume ≥1 m3) occurring in two subsequent years to be accurately predicted. In order to predict volume ranges of potential boulder falls, eight supervised machine learning algorithms are trained and validated based on sixty-five filtered samples with five represented features. Two key conclusions are drawn from the findings. First, the geological settings, altitudes, and gradients of source locations are demonstrated to be three triggering factors of different volume ranges of boulder falls. Second, the logistic regression algorithm has the best performance in predicting boulder fall volumes among all the algorithms. The f1-score of 5-fold cross-validation is 72.9%, and the prediction accuracy of the test dataset is 75.8%. We are confident that the methodologies and results in this paper will pave the way for analyzing other types of natural terrain hazards in addition to boulder falls in the long run. [ABSTRACT FROM AUTHOR]

This research concentrates on segmenting credit card clients of Taiwan into optimal groups. Unsupervised Learning plays a significant role in dividing customers into similar groups based on several parameters. If customers are clustered in groups optimally, it leads towards the retrieval of better precision from machine learning models applied to customers associated with the clusters. Different machine learning algorithms (Linear discriminant analysis, logistic regression and random forest) were applied on the obtained clusters through K-means, hierarchical and HK Means clustering algorithm, and predictive accuracy is compared with the accuracy obtained via applying mentioned machine learning models on the whole dataset. In this paper, a novel approach of combining K Means and hierarchical clustering (HK Means) is used. In this approach, HK means clustering algorithms are applied on the factorial coordinates, obtained from multiple correspondence analyses for segmenting customers into optimal groups has been proposed. The accuracy of the clustering techniques is evaluated from the decomposition of inertia. The results demonstrate that the combination of K Means and hierarchical clustering proved to be optimal clustering techniques for customer segmentation which can be used further for applying Machine Learning techniques for credit risk analysis. [ABSTRACT FROM AUTHOR]

Practical landslide predictions are instrumental to effective landslide risk management. Recently, the use of machine learning (ML) has become a promising alternative means for landslide predictions. This paper discusses the recent progress of a pilot study of ML-powered rainfall-based natural terrain landslide susceptibility analysis in Hong Kong. This study is different to other similar studies in that: (1) data sampling commonly used to deal with an imbalanced dataset is not adopted, and (2) the incorporation of domain knowledge on landslide characteristics for the development of physically meaningful ML models. The results are found to be promising, with the achieved ROC AUC up to 91.5% based on the testing data. The resolution of the susceptibility map is enhanced by approximately three orders of magnitude further than the introduction of additional features critically selected with feature engineering and based on domain knowledge and past experiences. [ABSTRACT FROM AUTHOR]

• This paper focused on the reliability of settlement prediction of immersed tunnel. • This paper focused on impact of time-dependent effects on settlement prediction. • A time-dependent expression for foundation modulus is proposed. • Several extreme conditions are considered in the future settlement prediction. The focus of immersed tunnel research has primarily revolved around settlement monitoring or prediction. However, there has been a lack of discussion regarding the time-dependent effects on settlement prediction, which could lead to obvious misjudgments in subsequent maintenance work. This paper addresses this issue by applying the physics-informed machine learning (PIML) algorithm to conduct an inverse analysis of the foundation modulus of the Hong Kong-Zhuhai-Macau Bridge (HZMB) tunnel, thereby identifying its variation tendency with time. The proposed time-dependent expression of foundation modulus demonstrates that the time-dependent effect is related to both post-construction settlement and the accumulation of back silting, which is significant and could persist throughout the entire service life of the HZMB tunnel. In further studies on settlement prediction for the HZMB tunnel, the results show that errors in prediction can exceed 30 mm at certain monitoring points. This error can be mitigated by considering the time-dependent effect, thereby significantly enhancing the reliability of predictions. [ABSTRACT FROM AUTHOR]

The dissemination of fake news on social media platforms is an issue of considerable interest, as it can be used to misinform people or lead them astray, which is particularly concerning when it comes to political events. The recent event of Hong Kong protests triggered an outburst of fake news posts that were identified on Twitter, which were then promptly removed and compiled into datasets to promote research. These datasets focusing on linguistic content were used in previous work to classify between tweets spreading fake and real news using traditional machine learning algorithms (Zervopoulos et al., in: IFIP international conference on artificial intelligence applications and innovations, Springer, Berlin, 2020). In this paper, the experimentation process on the previously constructed dataset is extended using deep learning algorithms along with a diverse set of input features, ranging from raw text to handcrafted features. Experiments showed that the deep learning algorithms outperformed the traditional approaches, reaching scores as high as 99.3% F1 Score, with the multilingual state-of-the-art model XLM-RoBERTa outperforming other algorithms using raw untranslated text. The combination of both traditional and deep learning algorithms allows for increased performance through the latter, while also gaining insight regarding tweet structure from the interpretability of the former. [ABSTRACT FROM AUTHOR]

Daylight data are required for energy-efficient building designs. However, daylight is scarcely measured, making the luminous efficacy model an alternative. This paper presents a method for modeling vertical luminous efficacy (Kvg) using measured data from measuring stations in Hong Kong. The artificial neural network (ANN), support vector machines (SVM), and empirical correlations were proposed for modeling Kvg. Machine learning (ML) models like ANN and SVM were used because they offer more accurate daylight predictions and ease in explaining complex relationships between atmospheric variables. Also, ML was explored since it has not been used in earlier vertical luminous efficacy studies. Sensitivity analysis was also carried out to determine the relative importance of input variables used for developing the proposed models. Findings show that scattering angle and diffuse fraction are crucial variables in vertical luminous efficacy modeling. Furthermore, when all proposed models were used to predict vertical daylight, it was observed that the peak relative root mean square error (%RMSE) was less than 18.6%. The obtained %RMSE showed that all models provided acceptable performance when evaluated against the measured daylight data. Finally, the findings also showed that the ANN models outperformed the SVM and empirical models. [ABSTRACT FROM AUTHOR]

Natural terrain landslides are mainly triggered by rainstorms in Hong Kong, which pose great threats to life and property. To mitigate landslide risk, building a prediction model which could provide information on both spatial and temporal probabilities of landslide occurrence is essential but challenging. In this paper, real-time rainfall conditions are incorporated into landslide prediction through a unique rainstorm-based database of reported landslides. Other landslide controlling factors related to topography, geology, and land cover are also considered. Five machine learning methods, including logistic regression, random forest, adaboost tree, support vector machine, and multilayer perceptron, are utilized and compared. Validated against historical rainstorms, the machine learning powered landslide prediction model could reasonably forecast the occurrence of landslides in a spatiotemporal context. Moreover, the effects of different rainstorm characteristics in terms of distinct rainfall spatial distribution and intensity on landslide susceptibility could also be captured by this model. For the landslide controlling factors investigated, rolling rainfall factors are proven to play a more important role than antecedent rainfall factors for landslide prediction. Among the five machine learning methods, the random forest model yields the most promising results in terms of all performance indicators (i.e., classification accuracy, recall, precision, area under curve, and overall accuracy). [ABSTRACT FROM AUTHOR]

Visually impaired people are facing many difficulties in daily life because of the disability in visual sensation. In daily consumption, they are hard to perform the transactions solely. They need to be assisted by other people in daily life, especially in purchasing, since they cannot recognize the different products with similar shapes and sizes. Also, in processing payment, visually impaired people are hard to differentiate the denominations (values) of banknotes. And many derived problems, such as cheating and providing the improper products, are being occurred. In this paper, an intelligent recognition system has been developed for visually impaired people in Hong Kong. The application system can help the users recognize canned foods and differentiate the nominal values of the banknotes so as to complete the monetary transactions efficiently and effectively. The application can be installed in smartphones with voice and vibration. By applying neural networks with assistive technology, the accuracies of recognizing banknotes and canned food are over 95% and 75%, respectively. [ABSTRACT FROM AUTHOR]

The editorial introduces issue 21.1 of Technoetic Arts via a critical reflection on the artificial intelligence hype (AI hype) that emerged in 2022 and gives an overview of each of the issue's ten articles. The first four articles engage with new technologies from different positions in art and architecture. These articles include an exploration of the indexical function of images as a means for revealing cultural premises governing public space during the pandemic in Hong Kong, a neuroarchitectural perspective on immersive architectural environments, and an investigation of the link between the popularity of immersive art installations and the emotions these installations generate. The 'Cryptoart' section features an ethnographic study of the DADA digital art collective, shedding light on the role of non-fungible tokens (NFTs) in community building. Guest-edited by Professor Tanu Gupta, the Special Section 'Perspectives from Chandigarh' constitutes the second part of the journal issue. Based on contributions to the national conference on Contemporary Perspectives in English Language, Literature, and Cultural Studies held on 15–16 July 2022, at Chandigarh University in Punjab, India, the six articles reflect a search for meaningful existence within colonial, patriarchal and biopolitical structures that frame everyday practices of exclusion and oppression. Discussing works of literature and cinema from the European, Japanese and the US American canon, the articles contribute a distinctive perspective to the critical analysis of contemporary societies that are shaped by the idea that progress is technological invention. [ABSTRACT FROM AUTHOR]

Spatial extension of lands through rapid reclamation is attractive for congested coastal megacities, although reclamations might suffer from project delay and budget overrun, often due to encountering unforeseen ground conditions. To accelerate reclamations, accurate prediction of soil consolidation over a construction duration of multiple years is needed for reclaimed lands, which often contain spatially varying subsurface stratigraphy and soil parameters. This calls for a spatio-temporal analysis of consolidation with a sound understanding of subsurface stratigraphic alternations of fine/coarse-grained soils and spatial variability of consolidation parameters (e.g. permeability). In this study, a unified framework, capable of simultaneously modelling stratigraphic variation and spatial variability of soil properties through machine learning of limited site investigation data, is combined with the finite-element method and Monte Carlo simulation for spatio-temporal consolidation analysis of reclaimed lands. The proposed method is applied to a real reclamation project in Hong Kong. Results indicate that the proposed method can accurately characterise subsurface geological cross-sections and spatially varying soil permeability with quantified uncertainty. Ignorance of spatial variability of soil permeability may result in an underestimation of consolidation time and an overestimation of undrained shear strength gain, and thus pose significant risks to reclamation projects. [ABSTRACT FROM AUTHOR]

This paper reports a new knowledge-driven engineered feature learning approach in response to the Global AI Challenge for Building E&M Facilities held by the Electrical and Mechanical Service Department (EMSD) of the Hong Kong SAR. The results were awarded with a Grand Prize by the competition organizer. A dynamically engineered multi-modal feature learning (DEMMFL) method is proposed for predicting the cooling load of two office buildings. The DEMMFL model is estimated with the Lasso-ridge regression and compared with other well-known methods such as the Lasso. The novel approach applies control system knowledge to engineer useful features and explore load patterns for multi-mode modeling. Deep learning methods including LSTM, GRU, and AutoGluon are implemented for automated machine learning and tested in parallel to compare the performance of the proposed model with existing methods. The proposed model is demonstrated to predict long-term cooling load most accurately using engineered features from weather information only. • A new dynamic feature engineering method for office building cooling load prediction. • Slow and fast dynamic features engineered from available weather data only. • Control-knowledge based feature to learn on-hours dynamic responses. • Multi-modal approach with eight day-type features to learn occupancy loads. • Deep recurrent learning and AutoML methods compared to the proposed method. [ABSTRACT FROM AUTHOR]

Background: The misestimation of surgical risk is a serious threat to the lives of patients when implementing surgical risk calculator. Improving the accuracy of postoperative risk prediction has received much attention and many methods have been proposed to cope with this problem in the past decades. However, those linear approaches are inable to capture the non-linear interactions between risk factors, which have been proved to play an important role in the complex physiology of the human body, and thus may attenuate the performance of surgical risk calculators.Methods: In this paper, we presented a new surgical risk calculator based on a non-linear ensemble algorithm named Gradient Boosting Decision Tree (GBDT) model, and explored the corresponding pipeline to support it. In order to improve the practicability of our approach, we designed three different modes to deal with different data situations. Meanwhile, considering that one of the obstacles to clinical acceptance of surgical risk calculators was that the model was too complex to be used in practice, we reduced the number of input risk factors according to the importance of them in GBDT. In addition, we also built some baseline models and similar models to compare with our approach.Results: The data we used was three-year clinical data from Surgical Outcome Monitoring and Improvement Program (SOMIP) launched by the Hospital Authority of Hong Kong. In all experiments our approach shows excellent performance, among which the best result of area under curve (AUC), Hosmer-Lemeshow test ([Formula: see text]) and brier score (BS) can reach 0.902, 7.398 and 0.047 respectively. After feature reduction, the best result of AUC, [Formula: see text] and BS of our approach can still be maintained at 0.894, 7.638 and 0.060, respectively. In addition, we also performed multiple groups of comparative experiments. The results show that our approach has a stable advantage in each evaluation indicator.Conclusions: The experimental results demonstrate that NL-SRC can not only improve the accuracy of predicting the surgical risk of patients, but also effectively capture important risk factors and their interactions. Meanwhile, it also has excellent performance on the mixed data from multiple surgical fields. [ABSTRACT FROM AUTHOR]

Representing building layouts as graphs can extract critical design patterns that would facilitate space syntax analyses as well as design mining and automation but traditional approaches (e.g., non-weighted adjacent graphs) encountered problems in modular buildings, as they are largely shaped under the principle of 'modularity' rather than freeform cast in-situ elements. This paper attempts to develop a novel analytical tool called ModularGraph to represent modular building layouts (MBLs) as graphs considering their unique adjacency, connectivity, and conjoint relationships in a triumvirate. It does so by developing a prototype then applying it to 36 modular buildings for iteration, finetuning, and finalizing. It is found that ModularGraph can effectively translate heterogeneous forms of MBLs into unified graph-based representations with rich graphic and semantic information. This study not only contributes an innovative graph analytic tool for design pattern mining, but also lays a stepping stone towards generative AI for modular building design. • Propose a graph triumvirate structure for modular building layout representation. • Incorporate a weight mechanism to indicate the strength of spatial relationships for better semantic interpretations. • Perform a quantitative analysis on the layouts of 36 modular buildings in Hong Kong using the proposed framework. • Uncover patterns in modular building layout designs exemplified by ModularGraph. • Develop a machine learning model to learn knowledge from layout designs represented by ModularGraph. [ABSTRACT FROM AUTHOR]

Active school commuting makes a vital contribution to physical activity, thus improving the health and well-being for children and adolescents. The built environment is widely acknowledged as a factor that can affect travel behavior. However, few studies assess the influence of micro-level streetscape features on active school commuting using street view images. Additionally, existing research often presupposes that the relationship between built environment and walking to school is linear or generalized linear. Using data from the 2016 census and street view images in Hong Kong, this paper investigates the non-linear relationship between the micro-level streetscape features and propensity of walking to school with several machine learning methods. A refined evaluation of streetscape features is conducted through semantic segmentation, object detection and perceptual color analysis. The results show that the non-linear model facilitates a deeper understanding of the genuine relationship between streetscape features and propensity of walking to school. Grass view index, mean saturation of buildings and number of traffic lights have greater significance in predicting walking propensity. Moreover, some of the streetscape features have salient threshold effects, indicating that environmental interventions would only be effective within a specific range. These findings can provide nuanced and fine-grained guidance for building a walkable, children-friendly and sustainable city. • Machine learning techniques are applied to explore the non-linear relationship. • Streetscape features exhibit salient non-linear and threshold effects on walking propensity. • The hierarchy of priorities and most effective ranges for environmental intervention are revealed. [ABSTRACT FROM AUTHOR]

Multipath and non-light-of-sight (NLOS) reception greatly deteriorate the GNSS positioning accuracy in urban areas. Currently, most of the attempts using machine learning in this area focus on signal status prediction (LOS, multipath, NLOS). This paper exploits the capability of deep learning in multipath/NLOS mitigation. A new input feature, the single-differenced residual map is proposed, which has a high correlation with the user location and is very effective in multipath/NLOS mitigation. Combining the domain knowledge in GNSS, features of residual maps from different satellites are extracted by the proposed network and generate the heat map to indicate the user location. The proposed network can significantly improve the positioning accuracy of 84% of the epochs in the dense urban to 5-meter level. In addition, our network has a superior generalization ability, reducing the error of 90% of the epochs to 7 m level in a new scenario. • A deep learning network structure to directly estimate urban GNSS positioning. • A new GNSS feature, single-differenced residual maps, is proposed to improve the positioning accuracy of the neural network. • The proposed network is validated and tested using real GNSS raw data collected in urban canyons at Hong Kong. [ABSTRACT FROM AUTHOR]

Forecasting time series is crucial for financial research and decision-making in business. The non-linearity of stock market prices has a profound impact on global economic and financial sectors. This study focuses on modeling and forecasting the daily prices of key stock indices - MASI, CAC 40, DAX, FTSE 250, NASDAQ, and HKEX, representing the Moroccan, French, German, British, US, and Hong Kong markets, respectively. We compare the performance of machine learning models, including Long Short-Term Memory (LSTM), eXtreme Gradient Boosting (XGBoost), and the hybrid LSTM-XGBoost, and utilize the skforecast library for backtesting. Results show that the hybrid LSTM-XGBoost model, optimized using Grid Search (GS), outperforms other models, achieving high accuracy in forecasting daily prices. This contribution offers financial analysts and investors valuable insights, facilitating informed decision-making through precise forecasts of international stock prices. [ABSTRACT FROM AUTHOR]

This study proposes a novel, new ensemble model (NEM) designed to simulate the maximum water level increases caused by storm surges in a frequently cyclone‐affected coastal water of Hong Kong, China. The model relies on storm and water level data spanning 1978–2022. The NEM amalgamates three machine learning algorithms: Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and XGBoost (XGB), employing a stacking technique for integration. Six parameters, determined using the Random Forest and Recursive Feature Elimination algorithms (RF‐RFE), are used as input features for the NEM. These parameters are the nearest wind speed, gale distance, nearest air pressure, minimum distance, maximum pressure drop within 24 hr, and large wind radius. Model assessment results suggest that the NEM exhibits superior performance over RF, GBDT, and XGB, delivering high stability and precision. It reaches a coefficient of determination (R2) up to 0.95 and a mean absolute error (MAE) that fluctuates between 0.08 and 0.20 m for the test data set. An interpretability analysis conducted using the SHapley Additive exPlanations (SHAP) method shows that gale distance and nearest wind speed are the most significant features for predicting peak water level increases during storm surges. The results of this study could provide practical implications for predictive models concerning storm surges. These findings present essential tools for the mitigation of coastal disasters and the improvement of marine disaster warning systems. Plain Language Summary: This research introduces an innovative ensemble machine learning model that predicts maximum water level rises due to storm surges in Hong Kong's cyclone‐prone coastlines. The model blends three machine learning algorithms, namely Random Forest (RF), Gradient Boosting Decision Tree (GBDT), and XGBoost (XGB), using storm and surge data from 1978 to 2022. It employs six parameters, including nearest wind speed and gale distance, providing superior performance with high stability and precision. Using SHapley Additive exPlanations (SHAP) analysis, gale distance and wind speed emerged as key predictors. The NEM could enhance storm surge models, aiding in coastal disaster mitigation and improving maritime warning systems. Key Points: An ensemble learning model has been developed to predict the peak water level rise during storm surges in Hong Kong's coastal regionsThe ensemble approach outperforms standalone machine learning methods of Random Forest, Gradient Boosting Decision Tree, and XGBoostAn interpretability analysis reveals that gale distance (Gale_Dis), and the nearest wind speed (N_WS) are the most influential features for storm surge prediction [ABSTRACT FROM AUTHOR]

State-of-the-art methods using attention mechanism in Recurrent Neural Networks have shown exceptional performance targeting sequential predictions and classifications. We explore the attention mechanism in Long–Short-Term Memory (LSTM) network based stock price movement prediction. Our proposed model significantly enhances the LSTM prediction performance in the Hong Kong stock market. The attention LSTM (AttLSTM) model is compared with the LSTM model in Hong Kong stock movement prediction. Further parameter tuning results also demonstrate the effectiveness of the attention mechanism in LSTM-based prediction method. [ABSTRACT FROM AUTHOR]

In light of advancements in big data analysis and artificial intelligence (AI), there are emerging opportunities to harness these technologies to address perceptive biases. This study examines the potential perceptive biases that may arise when construction mediation is quasi-imposed on the disputing parties. This can happen when mediation attempts are stipulated in the construction contract or court-directed. It is argued that, under such circumstances, a negative perception might arise over whether a bona fide mediation can be realised. Concerns include the fairness and timeliness of the process, as well as the practice of opportunistic mediating behaviours. With data collected from practising construction mediation practitioners in Hong Kong, the constructs of Perceptions of Bona Fide Mediation, Quasi-Imposition, and Positive Mediation Outcomes were first developed. Applying partial least square structural equation modelling to the relationship frameworks of the constructs, it was found that quasi-imposition is not as damaging as envisaged as far as having a bona fide mediation and attaining positive mediation outcomes are concerned. Moreover, a negative perception of the fairness and timeliness of the quasi-imposed mediation would jeopardise the integrity of a bona fide mediation. In this regard, utilizing NLP and machine learning algorithms offers a pioneering AI-driven approach to informing mediating parties, as well as reminding mediators to uphold the fairness and timeliness of the process for the purposes of reaching positive mediation outcomes. [ABSTRACT FROM AUTHOR]

The importance of monitoring sea level in coastal zones becomes more and more obvious in the era of global climate change, because, in coastal zones, although satellite altimetry is an ideal tool in measuring sea level over open ocean, but its accuracy often decreases significantly at coast due to land contamination. Although the accuracy of waveform processing algorithms for synthetic aperture altimeters has been improved in the last decade, the computational speed is still not fast enough to meet the requirements of real-time processing, and the accuracy cannot meet the needs of nearshore areas within 1 km from the coast. To improve the efficiency and accuracy in the coastal zone, this study proposed an innovative waveform retracking scheme for the coastal zone based on a multiple optimization parabolic cylinder algorithm (MOPCA) integrated with machine learning algorithms such as recurrent neural network and Bayesian estimation. The algorithm was validated using 153-pass repeat cycle data from Sentinel-6 over Qianliyan Island and Hong Kong–Wanshan Archipelago. The computational speed of the proposed algorithm was four to five times faster than the current operational synthetic aperture radar (SAR) retracking algorithm, and its accuracy within 0–20 km from the island was comparable to the most popular SAMOSA+ algorithm, better than the official data product provided by Sentinel-6. Especially, the proposed algorithm demonstrates remarkable stability in the sense of proceeding speed. It maintains consistent performance, even when dealing with intricate wave patterns within a proximity of 1 km from the coast. The results showed that the proposed scheme greatly improved the quality of coastal altimetry waveform retracking. [ABSTRACT FROM AUTHOR]

Rain-induced man-made slope failures pose great threats to public safety as most man-made slopes are formed in densely populated areas. A critical step in managing landslide risks is to predict the time, locations and consequences of slope failures in future rainstorms. Based on comprehensive databases of in-service man-made slopes, rainstorms and landslides in Hong Kong during the past 35 years, a spatio-temporal landslide forecasting model for man-made slopes is developed in this study within a unified machine learning framework. With a storm-based data integration strategy and multiclass classification on landslide scales, the framework incorporates landslide time and consequences in landslide susceptibility mapping to successfully achieve spatio-temporal landslide forecasting. The machine learning-based landslide forecasting model is validated against historical landslide incidents both temporally and spatially and through a case study of the June 2008 storm; the model significantly outperforms the prevailing statistical rainfall–landslide correlations in terms of prediction accuracy. The model can predict the real-time evolution of probabilities, scales and spatial distribution of landslides during the progression of a rainstorm, which can never be achieved by statistical methods. It can serve as an essential module for state-of-the-art landslide risk assessment and early warning. [ABSTRACT FROM AUTHOR]

The Hong Kong and Macao Special Administrative Regions, situated within China's Guangdong–Hong Kong–Macao Greater Bay Area, significantly influence and are impacted by their air quality conditions. Rapid urbanization, high population density, and air pollution from diverse factors present challenges, making the health of the atmospheric environment in these regions a research focal point. This study offers three key contributions: (1) It applied an interpretable dynamic Bayesian network (DBN) to construct a dynamic causal model of air quality in Hong Kong and Macao, amidst complex, unstable, multi-dimensional, and uncertain factors over time. (2) It investigated the dynamic interaction between meteorology and air quality sub-networks, and both qualitatively and quantitatively identified, evaluated, and understood the causal relationships between air pollutants and their determinants. (3) It facilitated an online collaborative forecast of air pollutant concentrations, enabling pollution warnings. The findings proposed that a DBN-based dynamic causal model can effectively explain and manage complex atmospheric environmental systems in Hong Kong and Macao. This method offers crucial insights for decision-making and the management of atmospheric environments not only in these regions but also for neighboring cities and regions with similar geographical contexts. [ABSTRACT FROM AUTHOR]

Data mining is an approach to discover knowledge from large data. Pollutant forecasting is an important problem in the environmental sciences. This paper tries to use data mining methods to forecast fine particles (PM2.5) concentration level in Hong Kong Central, which is a famous business centre in Asia. There are several classification algorithms available in data mining, such as Artificial Neural Network (ANN) and Support Vector Machine (SVM). ANN and SVM are both machine learning algorithm used in variant area. This paper builds PM2.5 concentration level predictive models based on ANN and SVM by using R packages. The data set includes 2008-2011 period meteorological data and PM2.5 data. The PM2.5 concentration is divided into 2 levels: low and high. The critical point is 40μg/m³ (24 hours mean), which is based on the standard of US Environmental Protection Agency (EPA). The parameters of both models are selected by multiple cross validation. According to 100 times 10-fold cross validation, the testing accuracy of SVM is around 0.803∼0.820, which is much better than ANN whose accuracy is around 0.746∼0.793. [ABSTRACT FROM AUTHOR]

In the era of big data, data is increasingly driving the construction waste management (CWM) for minimizing the impacts on the environment and recycling construction materials. However, missing data, led by various information barriers, often undermines the decision-making and hinders effective CWM. This paper applies aggregated behavior-based Machine Learning (ML) methods to handling the project-level 'Missing Not At Random' (MNAR) data by using aggregated waste generation behaviors as a case study. First, we define a set of 821 waste generation behavioral features based on waste big data, then screen the indicative and decisive behaviors using automatic feature selection. Then, the most predictive ML method, trained via data of 2,451 construction projects in 2011–2016 in Hong Kong, is selected for handling the MNAR data. The experiments showed that the prediction of project missing data was satisfactory (validation F 1 = 0.87, test F 1 = 0.80). The contribution of this paper is to pinpoint the potential of waste big data in portraying project behaviors for more value-added applications, at the same time, to present a handling method for MNAR data that is automatic, fast, and low-cost from the CWM practitioner's perspective. [Display omitted] [ABSTRACT FROM AUTHOR]

• Predicting departure delay in parallel may cause ambiguity in decision making. • A novel hierarchical integrated machine learning model helps to improve prediction. • The historical data provided by an international airline was used for the study. • The model is the best alternative to regression and multiclass classification. Flight delays may propagate through the entire aviation network and are becoming an important research topic. This paper proposes a novel hierarchical integrated machine learning model for predicting flight departure delays and duration in series rather than in parallel to avoid ambiguity in decision making. The paper analyses the proposed model using various machine learning algorithms in combination with different sampling techniques. The highly noisy, unbalanced, dispersed, and skewed historical high dimensional data provided by an international airline operating in Hong Kong was used to demonstrate the practical application of the model. The result shows that for a 4-h forecast horizon, a constructive neural network machine learning algorithm with the Synthetic Minority Over Sampling Technique-Tomek Links (SMOTETomek) sampling technique was able to achieve better average balanced recall accuracies of 65.5%, 61.5%, 59% for classifying delay status and predicting delay duration at thresholds of 60 min and 30 min, respectively. Similarly, for minority labels, the precision-recall and area under the curve showed that the proposed model achieved better results of 32.44% and 35.14% compared to the parallel model of 26.43% and 21.02% for thresholds of 60 min and 30 min, respectively. The effect of different sampling techniques, sampling approaches, and estimation mechanisms on prediction performance is also studied. [ABSTRACT FROM AUTHOR]

Background: Accurately estimating elderly patients' rehospitalisation risk benefits clinical decisions and service planning. However, research in rehospitalisation and repeated hospitalisation yielded only models with modest performance, and the model performance deteriorates rapidly as the prediction timeframe expands beyond 28 days and for older participants. Methods: A temporal zero-inflated Poisson (tZIP) regression model was developed and validated retrospectively and prospectively. The data of the electronic health records (EHRs) contain cohorts (aged 60+) in a major public hospital in Hong Kong. Two temporal offset functions accounted for the associations between exposure time and parameters corresponding to the zero-inflated logistic component and the Poisson distribution's expected count. tZIP was externally validated with a retrospective cohort's rehospitalisation events up to 12 months after the discharge date. Subsequently, tZIP was validated prospectively after piloting its implementation at the study hospital. Patients discharged within the pilot period were tagged, and the proposed model's prediction of their rehospitalisation was verified monthly. Using a hybrid machine learning (ML) approach, the tZIP-based risk estimator's marginal effect on 28-day rehospitalisation was further validated, competing with other factors representing different post-acute and clinical statuses. Results: The tZIP prediction of rehospitalisation from 28 days to 365 days was achieved at above 80% discrimination accuracy retrospectively and prospectively in two out-of-sample cohorts. With a large margin, it outperformed the Cox proportional and linear models built with the same predictors. The hybrid ML revealed that the risk estimator's contribution to 28-day rehospitalisation outweighed other features relevant to service utilisation and clinical status. Conclusions: A novel rehospitalisation risk model was introduced, and its risk estimators, whose importance outweighed all other factors of diverse post-acute care and clinical conditions, were derived. The proposed approach relies on four easily accessible variables easily extracted from EHR. Thus, clinicians could visualise patients' rehospitalisation risk from 28 days to 365 days after discharge and screen high-risk older patients for follow-up care at the proper time. [ABSTRACT FROM AUTHOR]

As occupational rehabilitation services are part of the public medical and health services in Hong Kong, work-injured workers are treated along with other patients and are not considered a high priority for occupational rehabilitation services. The idea of a work trial arrangement in the private market occurred to meet the need for a more coordinated occupational rehabilitation practice. However, there is no clear service standard in private occupational rehabilitation services nor concrete suggestions on how to offer rehabilitation plans to injured workers. Electronic Health Records (EHRs) data can provide a foundation for developing a model to improve this situation. This project aims at using a machine-learning-based approach to enhance the traditional prediction of disability duration and rehabilitation plans for work-related injury and illness. To help patients and therapists to understand the machine learning result, we also developed an interactive dashboard to visualize machine learning results. The outcome is promising. Using the variational autoencoder, our system performed better in predicting disability duration. We have around 30% improvement compared with the human prediction error. We also proposed further development to construct a better system to manage the work injury case. [ABSTRACT FROM AUTHOR]

The paper studies the horizontal global, direct-beam and sky-diffuse solar irradiance data measured in Hong Kong from 2008 to 2013. A machine learning algorithm was employed to predict the horizontal sky-diffuse irradiance and conduct sensitivity analysis for the meteorological variables. Apart from the clearness index (horizontal global/extra atmospheric solar irradiance), we found that predictors including solar altitude, air temperature, cloud cover and visibility are also important in predicting the diffuse component. The mean absolute error ( MAE ) of the logistic regression using the aforementioned predictors was less than 21.5 W/m 2 and 30 W/m 2 for Hong Kong and Denver, USA, respectively. With the systematic recording of the five variables for more than 35 years, the proposed model would be appropriate to estimate of long-term diffuse solar radiation, study climate change and develope typical meteorological year in Hong Kong and places with similar climates. [ABSTRACT FROM AUTHOR]

Characterizing and forecasting coastal water quality and spatiotemporal evolution should be significant to coastal ecosystem management. However, high-quality modeling coastal water quality and their spatiotemporal evolutions is rather challenging due to complex dynamic mechanisms especially in a spatially and temporally heterogenous semi-enclosed bay. To this end, this study develops a framework incorporating machine learning (ML) algorithms and the Environmental Kuznets Curves (EKC) analysis to model, analyze and forecast the spatiotemporal variations of water quality indicators for different subzones and seasons in the semi-enclosed Tolo Harbour of Hong Kong. The application results indicate that the developed ML-based framework with an accuracy range of 0.672 ∼ 0.998 is well-suited in forecasting and understanding the coastal water evolution in a semi-enclosed harbour compared to conventional approach. Furthermore, the spatiotemporal characteristics of coastal water quality evolution in this semi-enclosed bay are analyzed and discussed for coastal hydro-environmental management. Moreover, the EKC analysis is also performed for determining the evolutions of essential water quality variables under 95% confidence interval of Hong Kong PCGDP projection and then implemented in the developed ML-based model for future prediction. [ABSTRACT FROM AUTHOR]

We conducted a statistical study and developed a machine learning model to triage COVID-19 patients affected during the height of the COVID-19 pandemic in Hong Kong based on their medical records and test results (features) collected during their hospitalization. The correlation between the values of these features is studied against discharge status and disease severity as a preliminary step to identify those features with a more pronounced effect on the patient outcome. Once identified, they constitute the inputs of four machine learning models, Decision Tree, Random Forest, Gradient and RUSBoosting, which predict both the Mortality and Severity associated with the disease. We test the accuracy of the models when the number of input features is varied, demonstrating their stability; i.e., the models are already highly predictive when run over a core set of (6) features. We show that Random Forest and Gradient Boosting classifiers are highly accurate in predicting patients' Mortality (average accuracy ∼99%) as well as categorize patients (average accuracy ∼91%) into four distinct risk classes (Severity of COVID-19 infection). Our methodical and broad approach combines statistical insights with various machine learning models, which paves the way forward in the AI-assisted triage and prognosis of COVID-19 cases, which is potentially generalizable to other seasonal flus. [ABSTRACT FROM AUTHOR]

This paper describes the prediction of settlements induced by urban area tunneling using five machine learning (ML) algorithms. The settlement database, which was collected from a subway tunnel project in Hong Kong, consisted of 253 settlement measurements and 32 settlement influencing factors. The Bayesian optimization-based hyperparameter tuning was applied to efficiently explore optimal combinations and to enhance prediction performance. The optimal hyperparameters were selected by considering the three-fold cross-validation (CV) result of training data. The performance of the developed model was evaluated by comparing the root mean squared error (RMSE), mean absolute error (MAE), and coefficient of determination (R2) values. The extreme gradient boosting algorithm demonstrated the highest prediction accuracy with RMSE, MAE, and R2 values of 1.606, 1.331, and 0.835, respectively. • Tunneling-induced surface settlements predicted with five machine learning algorithms. • Bayesian optimization demonstrated excellent hyperparameter tuning search ability. • Prediction performance of trained models validated using three-fold cross-validation. • Consistency of prediction results validated with acceptable residual errors. • XGB model achieved the highest prediction performance with RMSE of 1.606. [ABSTRACT FROM AUTHOR]

The worldwide spread of COVID-19 has caused significant damage to people's health and economics. Many works have leveraged machine learning models to facilitate the control and treatment of COVID-19. However, most of them focus on clinical medicine and few on understanding the spatial dynamics of the high-risk population for transmission of COVID-19 in real-world settings. This study aims to investigate the association between population features and COVID-19 transmission risk in Hong Kong, which can help guide the allocation of medical resources and the implementation of preventative measures to control the spread of the pandemic. First, we built machine learning models to predict the number of COVID-19 cases based on the population features of different tertiary planning units (TPUs). Then, we analyzed the distribution of cases and the prediction results to find specific characteristics of TPUs leading to large-scale outbreaks of COVID-19. We further evaluated the importance and influence of various population features on the prediction results using SHAP values to identify indicators for high-risk populations for COVID-19 transmission. The evaluation of COVID-19 cases and the TPU dataset in Hong Kong shows the effectiveness of the proposed methods. The top three most important indicators are identified as people in accommodation and food services, low income, and high population density. [ABSTRACT FROM AUTHOR]

Previous evidence suggests that temperature is associated with the number of emergency department (ED) visits. A predictive system for ED visits, which takes local temperature into account, is therefore needed. This study aimed to compare the predictive performance of various machine learning methods with traditional statistical methods based on temperature variables and develop a daily ED attendance rate predictive model for Hong Kong. We analyzed ED utilization among Hong Kong older adults in May to September from 2000 to 2016. A total of 103 potential predictors were derived from 1- to 14-day lag of ED attendance rate and meteorological and air quality indicators and 0-day lag of holiday indicator and month and day of week indicators. LASSO regression was used to identify the most predictive temperature variables. Decision tree regressor, support vector machine (SVM) regressor, and random forest regressor were trained on the selected optimal predictor combination. Deep neural network (DNN) and gated recurrent unit (GRU) models were performed on the extended predictor combination for the previous 14-day horizon. Maximum ambient temperature was identified as a better predictor in its own value than as an indicator defined by the cutoff. GRU achieved the best predictive accuracy. Deep learning methods, especially the GRU model, outperformed conventional machine learning methods and traditional statistical methods. [ABSTRACT FROM AUTHOR]

Integrating borehole and piezocone penetration test (CPTU) data in site characterization helps to achieve a more comprehensive understanding of ground conditions. However, soil types at CPTU and nearby borehole locations may not always be consistent. The presence of noisy data or thin layers will mislead the interpretation of CPTU data in soil type classification and soil property evaluation. This study proposes a coupled machine learning method to integrate the borehole and CPTU data under a rigorous Bayesian framework and to identify and separate the noisy CPTU data without subjective judgment, which contributes to more reliable soil classification and property evaluation. The borehole-reported soil type and CPTU data are treated as two types of evidence of the authentic soil type. A lateral transition of soil type from the CPTU location to the borehole location is allowed to capture the discrepancy of soil types. The proposed approach is applied to the marine site characterization of the Hong Kong-Zhuhai-Macao Bridge that crosses the Pearl River Estuary of China. The soil seams embedded in the dominant soil strata are successfully detected, producing a more reliable soil profile and interpreting more compatible soil properties with engineering practice. Additionally, the integration of borehole and CPTU data significantly reduces the stratification uncertainty in site characterization. [ABSTRACT FROM AUTHOR]

The public specialty ophthalmic clinics in Hong Kong, under the Hospital Authority, receive tens of thousands of referrals each year. Triaging these referrals incurs a significant workload for practitioners and the other clinical duties. It is well-established that Hong Kong is currently facing a shortage of healthcare workers. Thus a more efficient system in triaging will not only free up resources for better use but also improve the satisfaction of both practitioners and patients. Machine learning (ML) has been shown to improve the efficiency of various medical workflows, including triaging, by both reducing the workload and increasing accuracy in some cases. Despite a myriad of studies on medical artificial intelligence, there is no specific framework for a triaging algorithm in ophthalmology clinics. This study proposes a general framework for developing, deploying and evaluating an ML-based triaging algorithm in a clinical setting. Through literature review, this study identifies good practices in various facets of developing such a network and protocols for maintenance and evaluation of the impact concerning clinical utility and external validity out of the laboratory. We hope this framework, albeit not exhaustive, can act as a foundation to accelerate future pilot studies and deployments. [ABSTRACT FROM AUTHOR]

Abstract: In recent years, Gaussian process (GP) models have been popularly studied to solve hard machine learning problems. The models are important due to their flexible non-parametric modeling abilities using Mercer kernels and the Bayesian framework for probabilistic inference. In this paper, we propose a sparse GP regression (GPR) model for tourism demand forecasting in Hong Kong. The sparsification procedure of the GPR model not only decreases the computational complexity but also improves the generalization ability. We experiment the proposed model with monthly demand data that are relevant to Hong Kong’s tourism industry, and compare the performance of the sparse GPR model with those of various kernel-based models to show its effectiveness. The proposed sparse GPR model shows that its forecasting capability outperforms those of the ARMA model and the two state-of-the-art SVM models. [Copyright &y& Elsevier]

Abstract: Harmful algal blooms, which are considered a serious environmental problem nowadays, occur in coastal waters in many parts of the world. They cause acute ecological damage and ensuing economic losses, due to fish kills and shellfish poisoning as well as public health threats posed by toxic blooms. Recently, data-driven models including machine-learning (ML) techniques have been employed to mimic dynamics of algal blooms. One of the most important steps in the application of a ML technique is the selection of significant model input variables. In the present paper, we use two extensively used ML techniques, artificial neural networks (ANN) and genetic programming (GP) for selecting the significant input variables. The efficacy of these techniques is first demonstrated on a test problem with known dependence and then they are applied to a real-world case study of water quality data from Tolo Harbour, Hong Kong. These ML techniques overcome some of the limitations of the currently used techniques for input variable selection, a review of which is also presented. The interpretation of the weights of the trained ANN and the GP evolved equations demonstrate their ability to identify the ecologically significant variables precisely. The significant variables suggested by the ML techniques also indicate chlorophyll-a (Chl-a) itself to be the most significant input in predicting the algal blooms, suggesting an auto-regressive nature or persistence in the algal bloom dynamics, which may be related to the long flushing time in the semi-enclosed coastal waters. The study also confirms the previous understanding that the algal blooms in coastal waters of Hong Kong often occur with a life cycle of the order of 1–2 weeks. [Copyright &y& Elsevier]

[Display omitted] • The communal areas can have different levels of energy performance in the 15 types of public residential buildings under study. • Spatial design features can evidently affect the energy consumption in the communal areas. • The ability to harvest daylight serves as a key determinant of energy efficiency. • Both econometric and machine learning techniques are proved to be effective to generate reliable findings. In this paper, we construct a unique database for 1228 residential buildings in Hong Kong to investigate how the spatial features of these residential buildings affect the electricity consumption in the communal area. We choose Hong Kong for this analysis as the city owns a large number of standard-type residential buildings managed by the public institution, which could be affected strongly by environmental policies. Both the machine learning method, based on the Least Absolute Shrinkage and Selection Operator (LASSO), and econometric regressions are adopted to analyse the data. We first utilize the machine learning LASSO technique to identify the most relevant factors for the subsequent econometric analysis. Our results show that the electricity demand for relatively low consumption building types, such as Twin Tower, is 6% lower than that of the high consumption building types. Newly constructed buildings usually belong to the medium consumption types, with the estimated monthly electricity consumption per apartment in communal areas to be around 50.2 kWh on average in 2020. These findings shed light on the nexus between spatial features and energy use for complex buildings, potentially contributing to the better crafting of energy-saving policy and the improvement of residential building programmes. [ABSTRACT FROM AUTHOR]

Landslide susceptibility assessment is essential for regional landslide risk assessment and mitigation. Most past studies involved cell-based analysis that takes landslide incidents as geo-spatial points. Nevertheless, given that a landslide is a two-dimensional polygon on maps and a three-dimensional object in the real world, an object-wise assessment is more logical. Fusing with artificial intelligence (AI) techniques, this paper proposes a novel AI-powered object-based landslide susceptibility assessment method to address this issue. First, landslide and non-landslide objects are defined based on an optimal object size determined by statistics of historical landslides. Next, landslide and non-landslide samples are constructed by integrating geoenvironmental data layers derived from multi-source data. Subsequently, AI techniques are applied to learn susceptibility prediction based on the prepared samples. To illustrate the proposed method, a comprehensive case study of Hong Kong is conducted, in which six AI algorithms are evaluated including logistic regression (area under curve, AUC = 0.949), random forest (AUC = 0.951), LogitBoost (AUC = 0.958), convolutional neural network (CNN) (AUC = 0.966), bidirectional long short-term memory architecture of recurrent neural network (BiLSTM-RNN) (AUC = 0.966), and CNN-LSTM (AUC = 0.972), among which the BiLSTM-RNN and CNN-LSTM algorithms are applied in landslide susceptibility assessment for the first time. Results confirm that the proposed object-based method outperforms the traditional cell-based method significantly. Equally importantly, the case study produced the first set of AI-based territory-wide landslide susceptibility maps for Hong Kong. These maps can be used as a fundamental tool for quantifying natural terrain landslide risk and identifying susceptible zones where landslide mitigation measures may be needed. • A novel artificial intelligence and object-based landslide susceptibility method is proposed and validated. • The first set of territory-wide landslide susceptibility maps for Hong Kong is produced based on AI techniques. • CNN-LSTM and BiLSTM-RNN are successfully applied to landslide susceptibility analysis for the first time. • The proposed AI-powered object-based method outperforms the cell-based method significantly. [ABSTRACT FROM AUTHOR]