Search

• Addressing imbalanced data in ML poses challenges due to class disproportion. • In some imbalanced datasets, false negatives impact more than false positives. • This work introduces the MinFNR algorithm to minimize False Negative Rates (FNR). • The new algorithm strategically combines data, algorithmic, and hybrid approaches. Imbalanced datasets present a challenge in machine learning, especially in binary classification scenarios where one class significantly outweighs the other. This imbalance often leads to models favoring the majority class, resulting in inadequate predictions for the minority class, specifically in false negatives. In response to this issue, this work introduces the MinFNR ensemble algorithm, designed to minimize False Negative Rates (FNR) in imbalanced datasets. The new approach strategically combines data-level, algorithmic-level, and hybrid-level approaches to enhance overall predictive capabilities while minimizing computational resources using the Set Covering Problem (SCP) formulation. Through a comprehensive evaluation of diverse datasets, MinFNR consistently outperforms individual algorithms, showing its potential for applications where the cost of false negatives is substantial, such as fraud detection and medical diagnosis. This work also contributes to ongoing efforts to improve the reliability and effectiveness of machine learning algorithms in real imbalanced scenarios. [ABSTRACT FROM AUTHOR]

Precast ultra-high-performance concrete (UHPC) structures (PUSs) have gained increasing research and application interest in civil engineering owing to the combination of advanced construction materials and methods. UHPC joints are critical parts of PUSs; thus, an accurate prediction of their shear capacity (SC) is essential to ensure structural safety and reliability. However, existing equations for predicting SC have limited accuracy and applicability owing to their simplified assumptions and restricted input parameters. To address these challenges, this study used machine learning (ML) approaches to develop a unified and accurate predictive model for various types of UHPC joints. A well-curated database containing 218 UHPC joints with diverse types and configurations was established. Six ensemble algorithms and four traditional algorithms were employed to develop predictive models, and eight existing equations were compared for performance evaluation. Both correlation-based and SHAP-based feature selection methods were used to optimize the model accuracy. The ensemble algorithms demonstrated better performance than the traditional individual algorithms, with the gradient boosting machine (GBM) model ranked as the best ML model for SC. The ML model outperformed existing equations in all evaluated metrics, demonstrating its accuracy and robustness. Furthermore, Shapley Additive exPlanations (SHAP) analysis was employed to interpret the ML model, thereby providing insights into influential features and their relationships. These findings demonstrate the advantages of ML methods in predicting the SC of UHPC joints and provide valuable guidance for the structural design and research on PUSs. • An updated database consisting of 218 push-off tests of various types of UHPC joints is established. • Ensemble and individual ML models are developed to predict the shear capacity of UHPC joints. • Correlation-based and SHAP-based feature selection methods are used to optimize the ML models. • The best ML model is compared with the existing equations to highlight the advantages of the ML method. • The ML models are globally and individually interpreted using the SHAP analysis. [ABSTRACT FROM AUTHOR]

Recently, driven by advancements in the payload capacity and endurance of drones, the logistics industry has shown significant interest in drone Last-Mile logistics. Efficient routing are crucial scientific challenges in drone delivery problems. In this study, we address the routing problem in heterogeneous drone delivery, which involves a large drone transporting multiple small drones to sub-regions for parcel delivery, aiming to both reduce the drones' distance costs and improve customer satisfaction, termed HDDPBO To tackle the HDDPBO problem effectively, we propose a voting-based ensemble multi-objective genetic approach, named VEMOGA, in which an improved clustering algorithm is developed to divide customers into K clusters, enabling each drone to handle multiple parcel deliveries within a sub-region. In this way, it reduces the complexity of HDDPBO by transforming it into multiple sub-problems. Secondly, a multi-objective genetic approach with heuristic operators is proposed to explore high-quality solutions, in which customized crossover and mutation operators are designed in the genetic approach, and a voting-based ensemble algorithm is designed to robustly select the Pareto frontier with high-quality convergence and diversity. Extensive experiments are conducted on synthetic instances to evaluate the proposed algorithm, and the experimental results demonstrate superior performance compared to three other baselines. Additionally, a real-world instance has been scrutinized to ascertain the applicability of Last-Mile logistics, and sensitivity analyses of pivotal factors have been conducted and several managerial insights pertinent are given to the drone-based Last-Mile logistics. • A multi-objective model is proposed for heterogeneous drone logistics. • A voting-based ensemble multi-objective genetic algorithm is proposed. • The proposed algorithm achieves higher satisfaction and lower costs. [ABSTRACT FROM AUTHOR]

This study explores the optimization of shared-bike station locations, emphasizing the cycling role in low-carbon mobility and local economic revitalization. A unique aspect of this research is the incorporation of the retail activity index in determining station locations, land-use mix, transit proximity, and population and employment densities. A novel weighting method, combining the multi-criteria decision method "Criteria Importance Through Inter-criteria Correlation (CRITIC)" with Ensemble-based predictive algorithms (Random Forest, XGBoost, Gradient Boosting), was developed to evaluate these factors. Using a geographic information system-based spatial model, this study conducted a suitability analysis by applying these weights to assess and propose bike station sites. The findings from the Seoul case indicate a balanced land-use mix, a low number of transactions linked to cyclists, and a significant gap in multimodal transport integration. The weighting results revealed a high priority for transit proximity and retail activity in station placement. Suitability scores across districts revealed significant variability, suggesting the need for greater integration of cycling policies into urban planning. High-scoring districts, such as Gangnam-gu and Seocho-gu, offer insights for improvements in less suitable areas. This study advocates expanded cycling planning, focusing on enhancing access to public transit and retail areas. [ABSTRACT FROM AUTHOR]

Over the years, ensemble algorithms became a common solution for supervised learning tasks as they show more robustness and usually are more accurate. But for unsupervised learning and clustering in particular ensemble methods remain poorly researched. In this work we propose a group of ensemble clustering algorithms by exploiting existing ideas and improving them with modern hyperparameter tuning algorithms and quality measure choosing heuristics. Our proposed approach uses MASSCAH to choose base clustering algorithm and optimize its hyperparameters using SMAC. The proposed solution was experimentally compared to existing solutions and k-means algorithm, tuned by an expert. Experiments were conducted on 64 synthetic datasets from Gaussian distribution and on 20 real-world datasets. We used 4 cluster validity indices to evaluate the quality of resulting partitions. From the results of experiments we can conclude that the proposed solution performs comparable to k-means tuned by an expert and outperforms existing ensemble clustering solutions. Apart from that, the proposed solution can be used as an automatic solution for clustering. [ABSTRACT FROM AUTHOR]

This study aims to enhance the solar energy harvesting capabilities of Unmanned Aerial Vehicles (UAVs), with a focus on integrating solar power to improve overall energy harvesting systems. The proposed method combines two independent renewable systems to extract electricity from the environment. UAV wings equipped with solar panels capture solar energy, employing optimal power point tracking for increased efficiency. Simulation results utilize an ensemble machine learning algorithm, incorporating environmental variables and UAV data to predict solar power output. A comparative analysis involving various machine learning algorithms provides additional insights gleaned from the UAV dataset. [ABSTRACT FROM AUTHOR]

Abstract The variability of renewable energy resources, due to the characteristic weather fluctuations, introduces uncertainty in generation output that are greater than the conventional energy reserves the grid uses to deal with the relatively predictable uncertainties in demand. The high variability of renewable generation makes forecasting critical for optimal balancing and dispatch of generation plants in a smarter grid. The challenge is to improve the accuracy and the confidence level of forecasts at a reasonable computational cost. Ensemble methods such as random forest (RF) and extra trees (ET) are well suited for predicting stochastic photovoltaic (PV) generation output as they reduce variance and bias by combining several machine learning techniques while improving the stability; i.e. generalisation capabilities. This paper investigated the accuracy, stability and computational cost of RF and ET for predicting hourly PV generation output, and compared their performance with support vector regression (SVR), a supervised machine learning technique. All developed models have comparable predictive power and are equally applicable for predicting hourly PV output. Despite their comparable predictive power, ET outperformed RF and SVR in terms of computational cost. The stability and algorithmic efficiency of ETs make them an ideal candidate for wider deployment in PV output forecasting. Highlights • Developed machine learning models for PV generation prediction. • Compared the performance of support vector machines (SVM) with random forest (RF) and extra trees (ET). • All developed models have comparable predictive power and are equally applicable for predicting PV system's output. • RF and ET models can be used to as a variable selection tools. • ET has significantly lower computational cost. [ABSTRACT FROM AUTHOR]

Abstract Predictive analytics play an important role in the management of decentralised energy systems. Prediction models of uncontrolled variables (e.g., renewable energy sources generation, building energy consumption) are required to optimally manage electrical and thermal grids, making informed decisions and for fault detection and diagnosis. The paper presents a comprehensive study to compare tree-based ensemble machine learning models (random forest – RF and extra trees – ET), decision trees (DT) and support vector regression (SVR) to predict the useful hourly energy from a solar thermal collector system. The developed models were compared based on their generalisation ability (stability), accuracy and computational cost. It was found that RF and ET have comparable predictive power and are equally applicable for predicting useful solar thermal energy (USTE), with root mean square error (RMSE) values of 6.86 and 7.12 on the testing dataset, respectively. Amongst the studied algorithms, DT is the most computationally efficient method as it requires significantly less training time. However, it is less accurate (RMSE = 8.76) than RF and ET. The training time of SVR was 1287.80 ms, which was approximately three times higher than the ET training time. Highlights • Developed predictive modelling techniques for solar energy systems. • Compared performance of SVR with random forest, extra trees and decision trees. • Training time of SVR was three times higher than the ET (extra trees) training time. • RF and ET models can be used to as a variable selection tools. [ABSTRACT FROM AUTHOR]

• Four machine learning (ML) models were developed to predict the yield strength of neutron-irradiated (YS irradiation) ferritic and martensitic (F/M) steels. • XGBOOST model was the best predictor, followed by RF, GBOOST and SVR. • Models were able to predict 70 to 75% of the validation dataset's data points within one standard deviation. • Synthetic experiments captured trained XGBOOST model efficiency towards predicting the trend between irradiation parameters and YS irradiation. Due to their superior resistance to radiation-induced damage, Ferritic/Martensitic (F/M) steels are promising structural materials for advanced nuclear reactors. This study investigates machine learning (ML) methodologies to predict the yield strength of neutron-irradiated F/M steels. Popular ML algorithms, such as Random Forest (RF), Extreme Gradient Boosting (XGBOOST), Gradient Boosting (GBOOST), and Support Vector Regression (SVR), were trained on a experimental dataset to understand the relationship between the input variables (e.g., irradiation dose, irradiation temperature, tensile test condition, heat treatment conditions and steels composition) and the output variable (yield strength). Further, the SHapley Additive exPlanations (SHAP) algorithm was employed to obtain the importance hierarchy of the input variables for their selection. Post-training and testing ML algorithms, their performance was evaluated by assessing their ability to predict the unseen/validation dataset. Among the algorithms tested, XGBOOST demonstrated the highest performance in predicting the validation dataset, followed by RF, GBOOST, and SVR. Synthetic experiments show that the trained ML algorithms can capture the trends between the irradiation input variables and the yield strength. Overall, the trained ML algorithms overcame challenges such as data uncertainties, smaller and sparser datasets, and other complexities and predicted almost 70% to 75% of the datapoints in the unseen/validation dataset within one standard deviation. [Display omitted] [ABSTRACT FROM AUTHOR]

Energy prediction models are used in buildings as a performance evaluation engine in advanced control and optimisation, and in making informed decisions by facility managers and utilities for enhanced energy efficiency. Simplified and data-driven models are often the preferred option where pertinent information for detailed simulation are not available and where fast responses are required. We compared the performance of the widely-used feed-forward back-propagation artificial neural network (ANN) with random forest (RF), an ensemble-based method gaining popularity in prediction – for predicting the hourly HVAC energy consumption of a hotel in Madrid, Spain. Incorporating social parameters such as the numbers of guests marginally increased prediction accuracy in both cases. Overall, ANN performed marginally better than RF with root-mean-square error (RMSE) of 4.97 and 6.10 respectively. However, the ease of tuning and modelling with categorical variables offers ensemble-based algorithms an advantage for dealing with multi-dimensional complex data, typical in buildings. RF performs internal cross-validation (i.e. using out-of-bag samples) and only has a few tuning parameters. Both models have comparable predictive power and nearly equally applicable in building energy applications. [ABSTRACT FROM AUTHOR]

Railway operations are inherently complex and source of several problems. In particular, track geometry defects are one of the leading causes of train accidents in the United States. This paper presents a solution approach which entails the construction of an ensemble classifier to forecast the degradation of track geometry. Our classifier is constructed by solving the problem from three different perspectives: deterioration, regression and classification. We considered a different model from each perspective and our results show that using an ensemble method improves the predictive performance. [ABSTRACT FROM AUTHOR]

• The spatial distribution of groundwater in the MODS of the Tarim Basin was evaluated. • The ensemble algorithm could improve the prediction accuracy relative to a single algorithm. • The groundwater level could be predicted with much greater accuracy at depths of >8 m than <8 m. Owing to the lack of hydrological monitoring stations and groundwater observations, particularly for natural habitats, the distribution of groundwater along the Mountainous (alluvial fan)-Oasis-Desert System (MODS) in the middle reaches of the Tarim River has not yet been systematically investigated. The performance of ensemble algorithms with multi-source/multi-scale (30 m, 90 m, 250 m, 500 m, and 1000 m) remote sensing data to predict the groundwater levels of the MODS of the Tarim Basin should be investigated. To improve the current knowledge on the spatial distribution of groundwater in this region, we conducted a study on the applicability of ensemble learning algorithms for groundwater assessments in typical regions of dryland, namely the Oasis-Taklimakan Desert (OTD), Nature land in Oasis system (NLOS), Irrigation area (IA), and Oasis system (OS). The results showed that the R2 values of the groundwater level prediction accuracy for the four ecosystems were 0.92, 0.96, 0.86, and 0.50, respectively, and their corresponding optimal ensemble algorithms were AR_GRBFN, AR_RF, RotationForest_MLP, and AR_ GRBFN. The scales corresponding to the optimal prediction accuracy of OTD, NLOS, IA, and OS were 250 m, 30 m, 250 m, and 90 m, respectively, and the effect of the scale on their respective groundwater level prediction accuracies (maximum value compared to minimum value) were 11.48%, 21.63%, 26.73%, and 10.59%. Random Subspace, Rotation Forest, and Additive Regression improved the prediction accuracy of the base learning method by 76%, 72%, and 64%, respectively, followed by Bagging, whereas Dagging greatly increased prediction errors. Rotation Forest and Random Subspace showed stable performances and guaranteed relatively low prediction errors. Among all ensemble algorithms, Additive Regression helped the base learners obtain relatively optimal prediction accuracies with high probabilities. Except for the OS system, groundwater level could be predicted with much greater accuracy at depths of >8 m than at <8 m. The contribution analysis showed that topography and land-use patterns controlled the spatial distribution of groundwater across MODS. The ensemble learning algorithm showed good performance using multi-source and multi-scale data. [ABSTRACT FROM AUTHOR]

In the last decade, injury statistics have not exhibited appreciable improvement within the construction industry, with current management strategies deployed reactively in response to typically safety lagging indicators (e.g., injury rates, lost workdays, and post-incident inspections). This situation suggests that proactive safety management approaches that rely on leading indicators for key decision-making need to be developed. To propel the third wave of construction safety management research, the current study aims at developing site risk models that generate predictions for safety risk leading indicators across different work zones and over the entire project lifecycles. Such leading indicators can be used to proactively anticipate worksite risks such that preventative measures are implemented in advance and/or adjusted in real-time as projects progress to dynamically monitor and enhance safety performance. The developed risk models are driven by ensemble machine learning algorithms trained using previous injury precursors and outcomes. The ensemble algorithms consider five base algorithms subsequently tuned and validated: naïve Bayes, decision trees, random forests, support vector machines, and artificial neural networks. A demonstration application is also presented herein, where the ensemble algorithms are employed to develop a risk model that forecasts leading indicators of site-specific risk levels, including financial implications of potential injuries and the most likely affected body parts. The results demonstrate that, unlike currently available approaches, machine learning-based site risk models may be able to proactively and dynamically assess and subsequently boost the safety performance of construction projects. Such site risk models are expected to provide project managers with key information to proactively prevent worksite accidents, prioritize their resources, enhance project safety culture, and minimize injury-related costs. • Develop project site risk models predicting safety risk leading indicators • Train ensemble machine learning algorithms on injury precursors and outcomes • Develop models to assess site safety risk levels proactively over time and space • Empower managers to prioritize resources for boosting project safety performance • Tools to predict and proactively intervene to mitigate workplace safety incidents [ABSTRACT FROM AUTHOR]

Compared to conventional alloys, multi-principal element alloys (MPEAs) offer attractive and tunable fatigue properties. With the advancement of machine learning (ML) in material science, it is perceived as an effective tool for predicting materials fatigue life. Therefore, in this work, we aim to predict the room temperature fatigue lives of two classes of MPEAs, the single-phase Co a Cr b Fe c Mn d Ni e system and the multi-phase Al f Co g Cr h Fe i Mn j Ni k system, using four different ML algorithms. Popular ML algorithms viz., Random Forest (RF), Support Vector Machine (SVM), and boosting algorithms such as GBOOST and XGBOOST were employed. Three pairs of test and training datasets were utilized to generalize the employed ML models. Input variables selection procedure was also carried out for possible improvement of algorithms' performance. SVM and boosting algorithms predicted fatigue lives of both classes of MPEAs within the band of the factor of two. Overall, all algorithms performed reasonably well, despite several hindering factors, such as the data's inherent scatter and limited datasets, which were acquired using different testing standards. [Display omitted] [ABSTRACT FROM AUTHOR]

• Ensemble algorithms are proposed for the bond strength prediction of the corroded reinforced concrete. • The models can reveal the nonlinear behavior of corroded reinforced concrete accurately. • The most crucial parameters affecting the bond strength are determined. Reinforced concrete bond strength deterioration is one of the most serious problems in the construction industry. It is one of the most common factors impacting structural deterioration and the major cause of premature decadence of reinforced concrete structures. Therefore, developing an accurate model with the lowest variance and high reliability for the bond strength of corroded reinforced concrete is very important. The current work evaluates the efficiency of convolution-based ensemble learning algorithms. To address these issues, convolution-based ensemble learning models are developed using a database collected from the previous experimental studies of relative bond strength for corroded reinforced concrete. Seven parameters are considered as inputs, including bar diameter, the ratio of concrete cover to bar diameter, the ratio of the bar diameter to embedment length, transverse reinforcement ratio, yielding strength, concrete compressive strength and the corrosion level. Results indicate that the convolution-based integrated stacking model produces excellent predictions with the coefficient of determination, a-20 index, and mean square error values of 0.84, 0.75, and 0.022. Since machine learning based models are commonly black-box models, the Shapley Additive Explanations (SHAP) method is used to explicitly relate the predicted output to the inputs. Based on SHAP, the corrosion level, concrete compressive strength, transverse reinforcement ratio, and the concrete cover to bar diameter ratio are ranked among the most influential parameters on the bond strength of the corroded reinforced concrete. [ABSTRACT FROM AUTHOR]

Building energy consumption prediction plays a key role in energy-efficiency decision making. With the advancement in data analytics, a number of machine learning-based building energy consumption prediction models have been developed in recent years. However, existing prediction models do not sufficiently take occupant behavior into account. Towards addressing this gap, this paper presents a machine-learning approach for predicting building energy consumption in an occupant-behavior-sensitive manner. In this approach, a model learns from a large set of energy-use cases that were modelled and simulated in EnergyPlus. The machine-learning prediction model was trained using a large dataset that includes 3-month hourly data for 5760 energy-use cases representing different combinations of building characteristics, outdoor weather conditions, and occupant behaviors. In developing the model, four machine-learning algorithms were tested and compared in terms of their prediction accuracy and computational efficiency: classification and regression trees (CART), ensemble bagging trees (EBT), artificial neural networks (ANN), and deep neural networks (DNN). The simulation results demonstrated the high impact of the variables considered in this study. For example, the highest energy-consuming case consumed over 3432 times more energy than the lowest-consuming case. Occupant behavior made a difference up to over 7 times in energy consumption. The DNN model with four hidden layers achieved 2.97% coefficient of variation (CV). Such high performance shows the potential of the proposed approach. The approach could help better understand the impact of occupant behavior on building energy consumption and identify opportunities for behavioral energy-saving measures. • An occupant-behavior-sensitive energy consumption prediction approach is presented. • Four machine-learning algorithms were tested and compared. • The machine-learning models learn from 5760 different energy-use cases. • The most accurate prediction model achieved CV of 2.97%. • The proposed approach could help better understand the impact of occupant behavior. [ABSTRACT FROM AUTHOR]