Search

• LULC changes of three decades from 1991 to 2021 were examined and predicted. • Multi-spectral remote sensing data and RFT were utilized to generate LULC maps. • ACR were considered to analyze the categorical land use change detection. • Impervious land has expanded 10 times within 30 years and will increase in the future. The integration of multi-spectral remote sensing and GIS-based analysis is significant for studying land cover changes, providing valuable insights for informed land management and sustainable development. The present study aims to examine land use land cover (LULC) changes of three decades from 1991 to 2021 and predict the future LULC change in Binh Duong province, Vietnam to explore a future research direction on land use change and associated challenges in the study region. Multi-spectral remote sensing data and random forest tree (RFT) were utilized to generate LULC maps. Areal statistics and annual change rate were considered to analyze the categorical land use change detection. Statistical measures such as user's accuracy, producer's accuracy, kappa coefficient, and confusion matrix were employed to assess the accuracy of LULC classification. To predict future LULC and simulate the spatio-temporal change, we considered previous year LULC maps, independent spatial variables and a combined artificial neural network (ANN) multi-layer perceptron approach. The analysis revealed that there was a huge transition from agricultural lands to residential land with industry and commerce which resulting an expansion of impervious lands and a rapid decline of agricultural land as well as of scrub land and barren lands, and a changeability of forest and plantation, croplands, and waterbodies. Our study revealed that the impervious land has expanded 10 times within 30 years and will increase in the future. [ABSTRACT FROM AUTHOR]

This study estimates both hourly and daily Downward Surface Solar Radiation (SSR) in Istanbul while determining the importance of variables on SSR using tree-based machine learning methods, namely Decision Tree (DT), Random Forest (RF), and Gradient Boosted Regression Tree (GBRT). The hourly and daily data of climatic factors for the period between January 2016 and December 2020 are gathered from the European Centre for Medium-Range Weather Forecasts' (ECMWF) ERA5 reanalysis data sets. In addition to the meteorology data, hourly data of selected aerosols are obtained from the Ministry of Environment, Urbanization and Climate Change. Temperature, cloud coverage, ozone level, precipitation, pressure, and two components of wind speeds, PM10, PM2.5, and SO2 are utilized to train and test the established models. The model performances are determined with the out-of-bag errors by calculating R-squared, MSE, RMSE, and MBE. The GBRT model is found to be the most accurate model with the lowest error rates. Furthermore, this study provides the variable importance in determining the SSR. Although all models provide different values for the variable importance; temperature, ozone level, cloud coverage, and precipitation are found to be the most important variables in estimating daily SSR. For the hourly estimation, the time of day (hour) becomes the most important factor in addition to temperature, ozone level, and cloud coverage. Finally, this study shows that the tree-based machine learning methods used with these variables to estimate hourly and daily SSR results are very accurate when it is not possible to measure the SSR values directly. [ABSTRACT FROM AUTHOR]

Effective winter road maintenance relies on precise road friction estimation. Machine learning (ML) models have shown significant promise in this; however, their inherent complexity makes understanding their inner workings challenging. This paper addresses this issue by conducting a comparative analysis of road friction estimation models using four ML methods, including regression tree, random forest, eXtreme Gradient Boosting (XGBoost), and support vector regression (SVR). We then employ the SHapley Additive exPlanations (SHAP) explainable artificial intelligence (AI) to enhance model interpretability. Our analysis on an Alberta dataset reveals that the XGBoost model performs best with an accuracy of 91.39%. The SHAP analysis illustrates the logical relationships between predictor features and friction within all three tree-based models, but it also uncovers inconsistencies within the SVR model, potentially attributed to insufficient feature interactions. Thus, this paper not only showcase the role of explainable AI in improving the ML interpretability of models for road friction estimation, but also provides practical insights that could improve winter road maintenance decisions. [ABSTRACT FROM AUTHOR]

Aims: To investigate high‐risk sociodemographic and environmental determinants of health (SEDH) potentially associated with adult obesity in counties in the United States using machine‐learning techniques. Materials and Methods: We performed a cross‐sectional analysis of county‐level adult obesity prevalence (body mass index ≥30 kg/m2) in the United States using data from the Diabetes Surveillance System 2017. We harvested 49 county‐level SEDH factors that were used in a classification and regression trees (CART) model to identify county‐level clusters. The CART model was validated using a 'hold‐out' set of counties and variable importance was evaluated using Random Forest. Results: Overall, we analysed 2752 counties in the United States, identifying a national median (interquartile range) obesity prevalence of 34.1% (30.2%, 37.7%). The CART method identified 11 clusters with a 60.8% relative increase in prevalence across the spectrum. Additionally, seven key SEDH variables were identified by CART to guide the categorization of clusters, including Physically Inactive (%), Diabetes (%), Severe Housing Problems (%), Food Insecurity (%), Uninsured (%), Population over 65 years (%) and Non‐Hispanic Black (%). Conclusion: There is significant county‐level geographical variation in obesity prevalence in the United States, which can in part be explained by complex SEDH factors. The use of machine‐learning techniques to analyse these factors can provide valuable insights into the importance of these upstream determinants of obesity and, therefore, aid in the development of geo‐specific strategic interventions and optimize resource allocation to help battle the obesity pandemic. [ABSTRACT FROM AUTHOR]

To accurately and rapidly predict seismic responses, including the maximum displacement (MaxD) and maximum acceleration (MaxA), of the isolated structure considering the soil–structure interaction (SSI), five ensemble learning models, i.e. random forest (RF), gradient boosting regression tree (GBRT), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM) and stacking model, are constructed. Firstly, a total of 96 000 nonlinear time history analyses of the isolated structure considering the SSI are conducted with the aid of OpenSees. The generated database is used for training and testing ensemble learning models. The ensemble learning models have 12 input variables in four categories, i.e. ground motion parameters, structural parameter, isolation parameters and soil parameter, and two output variables, i.e. MaxD and MaxA. The study shows that all ensemble learning models have excellent prediction performance for both training and testing datasets. The determination coefficients are larger than 0.96 and root-mean-square errors (RMSEs) are relatively small. Among the five ensemble learning models, the stacking model exhibits the best performance. In addition, the calculation method of feature importance score for the stacking model is provided. According to the feature importance analysis, the ground motion parameters have greater impact on seismic responses than other three categories of inputs. Finally, six ground motions are randomly selected to verify the generalization ability of the proposed ensemble learning models. The results show that the stacking model has a favorable generalization ability with relatively small prediction errors. [ABSTRACT FROM AUTHOR]

Many individual studies in the literature observed the superiority of tree-based machine learning (ML) algorithms. However, the current body of literature lacks statistical validation of this superiority. This study addresses this gap by employing five ML algorithms on 200 open-access datasets from a wide range of research contexts to statistically confirm the superiority of tree-based ML algorithms over their counterparts. Specifically, it examines two tree-based ML (Decision tree and Random forest) and three non-tree-based ML (Support vector machine, Logistic regression and k-nearest neighbour) algorithms. Results from paired-sample t-tests show that both tree-based ML algorithms reveal better performance than each non-tree-based ML algorithm for the four ML performance measures (accuracy, precision, recall and F1 score) considered in this study, each at p<0.001 significance level. This performance superiority is consistent across both the model development and test phases. This study also used paired-sample t-tests for the subsets of the research datasets from disease prediction (66) and university-ranking (50) research contexts for further validation. The observed superiority of the tree-based ML algorithms remains valid for these subsets. Tree-based ML algorithms significantly outperformed non-tree-based algorithms for these two research contexts for all four performance measures. We discuss the research implications of these findings in detail in this article. [ABSTRACT FROM AUTHOR]

Short-term load forecasting (STLF) plays a vital role in ensuring the safe, efficient, and economical operation of power systems. Accurate load forecasting provides numerous benefits for power suppliers, such as cost reduction, increased reliability, and informed decision-making. However, STLF is a complex task due to various factors, including non-linear trends, multiple seasonality, variable variance, and significant random interruptions in electricity demand time series. To address these challenges, advanced techniques and models are required. This study focuses on the development of an efficient short-term power load forecasting model using the random forest (RF) algorithm. RF combines regression trees through bagging and random subspace techniques to improve prediction accuracy and reduce model variability. The algorithm constructs a forest of trees using bootstrap samples and selects random feature subsets at each node to enhance diversity. Hyperparameters such as the number of trees, minimum sample leaf size, and maximum features for each split are tuned to optimize forecasting results. The proposed model was tested using historical hourly load data from four transformer substations supplying different campus areas of the University of Beira Interior, Portugal. The training data were from January 2018 to December 2021, while the data from 2022 were used for testing. The results demonstrate the effectiveness of the RF model in forecasting short-term hourly and one day ahead load and its potential to enhance decision-making processes in smart grid operations. [ABSTRACT FROM AUTHOR]

Acidification is a well-known treatment to reduce NH 3 emissions from livestock slurries by lowering their pH, but its application at the farm scale is still limited. The acid dosage depends on the acid strength and slurries composition. Acidification does not have a lasting effect and after the acid addition the slurry pH tends to rise due to its buffer system. By studying 54 samples of pig slurry, dairy cattle slurry and digestate, this study aimed to: (i) identify the chemical-physical parameters related to the sulphuric acid (98% w/w) dosage necessary to reduce pH at 5.5, and pH variation over time; (ii) develop predictive models for the acid dosage and the pH after one (pH1w) and two weeks (pH2w) of storage based on slurry composition by using regression trees (RT) and random forests (RF). Acid dosage ranged between 0.8 and 11.7 ml kg−1 increasing with slurry alkalinity, with digestate requiring significantly higher dosage than slurries. Pig slurry showed significantly higher pH increase than the other two slurries. Finally, the pH trend over time was negatively correlated with the solids content. The RF identified the alkalinity and the initial slurry pH as the most important variables in explaining the required acid dosage, while for pH1w and pH2w it identified the total organic carbon and volatile solids. Based on RF results, RT models accurately predicted required acid dosage (r2 = 0.881), the pH1w (r2 = 0.728) and pH2w (r2 = 0.667). Therefore, these simple models can have practical applications for reducing NH 3 emissions. • Acidification treatment reduces ammonia emissions from livestock slurry and digestate. • Correlations among slurry composition variables, acid dosage and pH trends were studied. • Needed acid dosage and pH trends were predicted with regression trees and random forests. • Needed acid dosage was accurately predicted by alkalinity and initial pH of the slurry. • PH trends over time were accurately predicted by total organic carbon and volatile solids. [ABSTRACT FROM AUTHOR]

The prognosis of wind turbine failures in real operating conditions is a significant gap in the academic literature and is essential for achieving viable performance parameters for the operation and maintenance of these machines, especially those located offshore. This paper presents a framework for estimating the remaining useful life (RUL) of the main bearing using regression models fed operational data (temperature, wind speed, and the active power of the network) collected by a supervisory control and data acquisition (SCADA) system. The framework begins with a careful data filtering process, followed by creating a degradation profile based on identifying the behavior of temperature time series. It also uses a cross-validation strategy to mitigate data scarcity and increase model robustness by combining subsets of data from different available turbines. Support vector, gradient boosting, random forest, and extra trees models were created, which, in the tests, showed an average of 20 days in estimating the remaining useful life and presented mean absolute error (MAE) values of 0.047 and mean squared errors (MSE) of 0.012. As its main contributions, this work proposes (i) a robust and effective regression modeling method for estimating RUL based on temperature and (ii) an approach for dealing with a lack of data, a common problem in wind turbine operation. The results demonstrate the potential of using these forecasts to support the decision making of the teams responsible for operating and maintaining wind farms. [ABSTRACT FROM AUTHOR]

Identifying colour from a distance is challenging due to the external noise associated with the measurement process. The present study focuses on developing a colour measuring system and a novel Multi-target Regression (MTR) model for accurate colour measurement from distance. Herein, a novel MTR method, referred as Multi-Objective Stacked Regression (MOSR) is proposed. The core idea behind MOSR is based on stacking as an ensemble approach with multi-objective evolutionary learning using NSGA-II. A multi-objective optimization approach is used for selecting base learners that maximises prediction accuracy while minimising ensemble complexity, which is further compared with six state-of-the-art methods over the colour dataset. Classification and regression tree (CART), Random Forest (RF) and Support Vector Machine (SVM) were used as regressor algorithms. MOSR outperformed all compared methods with the highest coefficient of determination values for all three targets of the colour dataset. Rigorous comparison with state-of-the-art methods over 18 benchmarked datasets showed MOSR outperformed in 15 datasets when CART was used as a regressor algorithm and 11 datasets when RF and SVM were used as regressor algorithms. The MOSR method was statistically superior to compared methods and can be effectively used to measure accurate colour values in the distance-based colour measuring device. [ABSTRACT FROM AUTHOR]

This article shows that decision trees constructed with Classification and Regression Trees (CART) and C4.5 methodology are consistent for regression and classification tasks, even when the number of predictor variables grows sub-exponentially with the sample size, under natural 0-norm and 1-norm sparsity constraints. The theory applies to a wide range of models, including (ordinary or logistic) additive regression models with component functions that are continuous, of bounded variation, or, more generally, Borel measurable. Consistency holds for arbitrary joint distributions of the predictor variables, thereby accommodating continuous, discrete, and/or dependent data. Finally, we show that these qualitative properties of individual trees are inherited by Breiman's random forests. A key step in the analysis is the establishment of an oracle inequality, which allows for a precise characterization of the goodness of fit and complexity tradeoff for a mis-specified model. for this article are available online. [ABSTRACT FROM AUTHOR]

Context: As evaluation of carcass components is costly and time consuming, models for prediction of broiler carcass components are useful. Aims: The aim was to investigate the feasibility of machine learning methods in the prediction of carcass components from measurements on live birds during the rearing period. Methods: Three machine learning methods, including regression tree, random forest and gradient-boosting trees, were applied to predict carcass yields, and benchmarked against classical linear regression. Two scenarios were defined for prediction. In the first scenario, carcass yields were predicted by live bodyweight, shank length and shank diameter features, recorded at 2, 3 and 4 weeks of age. In the second scenario, predictor features recorded at 5, 6 and 7 weeks of age were used. The two scenarios were reanalysed by including effective single-nucleotide polymorphisms associated with bodyweight, shank length and shank diameter as new predictor features. Key results: The correlation coefficient between predicted and observed values for predicting weight of carcass traits ranged from 0.50 for wing to 0.59 for thigh in the first scenario, and from 0.63 for wing to 0.74 for carcass in the second scenario. These predictions for the percentage of carcass components ranged from 0.30 for wing to 0.39 for carcass and breast in the first scenario, and from 0.34 for thigh to 0.43 for carcass in the second scenario when random forest was used. Conclusions: Predictive accuracy in the first scenario was lower than in the second scenario for all prediction methods. Including single-nucleotide polymorphisms as predictor features in either scenario did not increase the accuracy of the prediction. Implications: In general, random forest had the best performance among machine learning methods, and classical linear regression in two scenarios, suggesting that it may be considered as an alternative to conventional linear models for prediction of carcass traits in broiler chickens. Evaluation of carcass component yields is costly and time consuming, thus models for the prediction of broiler carcass yields are useful. Three advanced predictive models were applied to predict carcass component yields, and benchmarked against the classical linear regression. Random forest had the best performance among machine learning methods, suggesting that it may be considered as an alternative to conventional linear models for prediction of carcass yields in broiler chickens. In all of the predictive model tested in this study, inclusion of single-nucleotide polymorphisms as predictor features lowered the predictive performance of the model. [ABSTRACT FROM AUTHOR]

Post-stratification is applied when the subpopulation membership is observed only for sampled values and the goal is to estimate stratum-specific parameters which leads the survey statisticians towards primary goals i.e., classification of non-sampled units into different strata and prediction of the values of the study variables. Regression models, on one side, optimize the prediction of the study variable's non-sampled values while the classification algorithms, on the other side, look for the classification of non-sampled cases into different strata. Hence, it is crucial to deal with these two goals simultaneously for the estimation of stratum-specific parameters. This study introduces the idea of a double-objective classification and regression trees (CARTs) approach for estimating stratum-specific parameters. Theoretical properties of the total estimator are derived. An application on the estimation of health outcomes in different domains is given to delineate the practical significance as well as the efficiency of the proposed CART-based method. The proposed estimator of population total performs better than the existing stratum-specific estimator in terms of relative efficiency for all choices of parameters. As an ensemble model, the random forest CART outperforms the other competing tree-based models and homogenous population model without using any auxiliary variable. [ABSTRACT FROM AUTHOR]

The prediction of flight delays is one of the important and challenging issues in the field of scheduling and planning flights by airports and airlines. Therefore, in recent years, we have witnessed various methods to solve this problem using machine learning techniques. In this article, a new method is proposed to address these issues. In the proposed method, a group of potential indicators related to flight delay is introduced, and a combination of ANOVA and the Forward Sequential Feature Selection (FSFS) algorithm is used to determine the most influential indicators on flight delays. To overcome the challenges related to large flight data volumes, a clustering strategy based on the DBSCAN algorithm is employed. In this approach, samples are clustered into similar groups, and a separate learning model is used to predict flight delays for each group. This strategy allows the problem to be decomposed into smaller sub-problems, leading to improved prediction system performance in terms of accuracy (by 2.49%) and processing speed (by 39.17%). The learning model used in each cluster is a novel structure based on a random forest, where each tree component is optimized and weighted using the Coyote Optimization Algorithm (COA). Optimizing the structure of each tree component and assigning weighted values to them results in a minimum 5.3% increase in accuracy compared to the conventional random forest model. The performance of the proposed method in predicting flight delays is tested and compared with previous research. The findings demonstrate that the proposed approach achieves an average accuracy of 97.2% which indicates a 4.7% improvement compared to previous efforts. [ABSTRACT FROM AUTHOR]

The conventional method for the color-matching process involves the compounding of polymers with pigments and then preparing plaques by using injection molding before measuring the color by an offline spectrophotometer. If the color fails to meet the L*, a*, and b* standards, the color-matching process must be repeated. In this study, the aim is to develop a machine learning model that is capable of predicting offline color using data from inline color measurements, thereby significantly reducing the time that is required for the color-matching process. The inline color data were measured using an inline process spectrophotometer, while the offline color data were measured using a bench-top spectrophotometer. The results showed that the Bagging with Decision Tree Regression and Random Forest Regression can predict the offline color data with aggregated color differences (dE) of 10.87 and 10.75. Compared to other machine learning methods, Bagging with Decision Tree Regression and Random Forest Regression excel due to their robustness, ability to handle nonlinear relationships, and provision of insights into feature importance. This study offers valuable guidance for achieving Bagging with Decision Tree Regression and Random Forest Regression to correlate inline and offline color data, potentially reducing time and material waste in color matching. Furthermore, it facilitates timely corrections in the event of color discrepancies being observed via inline measurements. [ABSTRACT FROM AUTHOR]

Snow cover information is essential for pursuing seasonal variation studies in Himalayan river basins. This study aims to investigate the seasonal variation of snow cover in the Sutlej river basin (Tibet to Bhakra dam in India) over three different seasons: Monsoon (June–September), winter (October-January), and summer (February-May) during the period 2010–2021. Landsat 7 and 8 Surface Reflectance (SR) data is used to develop 108 land use land cover (LULC) maps for 12 years, with three seasons per year and three machine-learning models. The study has conducted on the Google Earth Engine (GEE) platform, employing Random Forest (RF), Classification and Regression Trees (CART), and Support Vector Machine (SVM) models to classify the Landsat satellite data and assess the seasonal snow cover variation during the three seasons. The results show that among the three machine learning models, the RF model exhibits the highest average overall accuracy at 98.75%, followed by the CART model at 98.10%, and the SVM model with the lowest accuracy at 97.15%. In terms of snow cover area variability; there is a decline trend in summer snow cover and an increase in monsoon snow cover over the past three years (2019–2021). However, an increasing trend emerges when considering the decadal changes in all three seasons. In addition, the maximum percentages of snow cover area observed as 67.61%, 46.78%, and 30.58% in the summer period of 2013, the winter period of 2019, and the monsoon period of 2021, respectively. Similarly, the minimum percentage of snow cover is 23.22%, 11.36%, and 13.01%, observed in the summer period of 2014, the winter period of 2011, and the monsoon period of 2012, respectively. A comprehensive assessment procedure on temporal and seasonal snow cover variation in a large river basin have been presented in this work, which will help to plan and manage the sustainable water resources in study region. [ABSTRACT FROM AUTHOR]

Ensemble methods, such as Bagging, Boosting, or Random Forests, often enhance the prediction performance of single learners on both classification and regression tasks. In the context of regression, we propose a gradient boosting‐based algorithm incorporating a diversity term with the aim of constructing different learners that enrich the ensemble while achieving a trade‐off of some individual optimality for global enhancement. Verifying the hypotheses of Biau and Cadre's theorem (2021, Advances in contemporary statistics and econometrics—Festschrift in honour of Christine Thomas‐Agnan, Springer), we present a convergence result ensuring that the associated optimization strategy reaches the global optimum. In the experiments, we consider a variety of different base learners with increasing complexity: stumps, regression trees, Purely Random Forests, and Breiman's Random Forests. Finally, we consider simulated and benchmark datasets and a real‐world electricity demand dataset to show, by means of numerical experiments, the suitability of our procedure by examining the behavior not only of the final or the aggregated predictor but also of the whole generated sequence. [ABSTRACT FROM AUTHOR]

Rental bike sharing is an urban mobility model that is affordable and ecofriendly. The public bike sharing model is widely used in several cities across the world over the past decade. Because bike use is rising constantly, understanding the system demand in prediction is important to boost the operating system readiness. This article presents a prediction model to meet user demands and efficient operations for rental bikes using Random Forest (RF), which is a homogeneous ensemble method. The approach is carried out in Seoul, South Korea to predict the hourly use of rental bikes. RF is compared with Support Vector Machine with Radial Basis Function Kernel, k‐nearest neighbor and Classification and Regression Trees to verify RF supremacy in rental bike demand prediction. Performance Index measures the efficiency of RF compared to the other predictive models. Also, the variable importance analysis is performed to assess the most important characteristics during different seasons by creating a predictive model using RF for each season. The results show that the influence of variables changes depending on the seasons that suggest different operating conditions. RF models trained with yearly and seasonwise models show that bike sharing demand can be further improved by considering seasonal change. [ABSTRACT FROM AUTHOR]

Pan evaporation modeling and forecasting are needed to provide timely, continuous, and valuable information to support water management. This study aimed to overcome the constraints identified in traditional regression techniques and less explored machine learning models—the CatBoost, to enhance the precision and comprehensiveness of pan evaporation estimation. Additionally, the CatBoost model was compared with other machine-learning approaches, viz., multiple linear regression (MLR), multiple non-linear least-squares regression (MNLSR), multivariate adaptive regression splines (MARS), random forest regression (RF), and M5 model tree (M5Tree). The algorithms were developed using data obtained from the Crop Research Center, Pantnagar- Uttarakhand, India. Stepwise regression was used as an input variable selection algorithm for selecting the best relevant input from several meteorological variables. For the model's development, data were split into two subsets: the first three years were reserved for model calibration, while the fourth-year data was used for model validation. Statistical analyses, namely root mean square error (RMSE), Nash–Sutcliffe efficiency (NSE), Willmott index of agreement (WI), Pearson correlation coefficient (PCC), coefficient of determination (R2), and mean absolute error (MAE) were used to assess the effectiveness of weekly pan evaporation estimation models. Findings indicated that the CatBoost model, based on four predictors, was the most accurate at Pantnagar station, with NSE, WI, MAE, RMSE, PCC and R2 ranging from 0.9838 to 0.9877, 0.9960–0.9968, 0.1991–0.2078, 0.3643–0.3650–0.9955, 0.9925 and 0.9851–0.9911, respectively. RF model was found to be the second-most best-performing model in terms of in accuracy and effectiveness for mapping weekly pan evaporation, followed by MARS, MNLSR, MLR, and M5Tree. In conclusion, CatBoost's approach of considering combinations greedily implies a thoughtful and cautious selection of features during the modeling process. This can enhance the model's robustness and generalization capability, contributing to more accurate predictions in diverse environmental conditions in subtropical and sub-humid regions. [ABSTRACT FROM AUTHOR]

Coronavirus disease 2019 (COVID-19) has strained healthcare systems worldwide. Predicting COVID-19 severity could optimize resource allocation, like oxygen devices and intensive care. If machine learning model could forecast the severity of COVID-19 patients, hospital resource allocation would be more comfortable. This study evaluated machine learning models using electronic records from 3,996 COVID-19 patients to forecast mild, moderate, or severe disease up to 2 days in advance. A deep neural network (DNN) model achieved 91.8% accuracy, 0.96 AUROC, and 0.90 AUPRC for 2-day predictions, regardless of disease phase. Tree-based models like random forest achieved slightly better metrics (random forest: 94.1% of accuracy, 0.98 AUROC, 0.95 AUPRC; Gradient boost: 94.1% of accuracy, 0.98 AUROC, 0.94 AUPRC), prioritizing treatment factors like steroid use. However, the DNN relied more on fixed patient factors like demographics and symptoms in aspect to SHAP value importance. Since treatment patterns vary between hospitals, the DNN may be more generalizable than tree-based models (random forest, gradient boost model). The results demonstrate accurate short-term forecasting of COVID-19 severity using routine clinical data. DNN models may balance predictive performance and generalizability better than other methods. Severity predictions by machine learning model could facilitate resource planning, like ICU arrangement and oxygen devices. [ABSTRACT FROM AUTHOR]

What dataset features affect machine learning (ML) performance has primarily been unknown in the current literature. This study examines the impact of tabular datasets' different meta-level and statistical features on the performance of various ML algorithms. The three meta-level features this study considered are the dataset size, the number of attributes and the ratio between the positive (class 1) and negative (class 0) class instances. It considered four statistical features for each dataset: mean, standard deviation, skewness and kurtosis. After applying the required scaling, this study averaged (uniform and weighted) each dataset's different attributes to quantify its four statistical features. We analysed 200 open-access tabular datasets from the Kaggle (147) and UCI Machine Learning Repository (53) and developed ML classification models (through classification implementation and hyperparameter tuning) for each dataset. Then, this study developed multiple regression models to explore the impact of dataset features on ML performance. We found that kurtosis has a statistically significant negative effect on the accuracy of the three non-tree-based ML algorithms of the Support vector machine (SVM), Logistic regression (LR) and K-nearest neighbour (KNN) for their classical implementation with both uniform and weighted aggregations. This study observed similar findings in most cases for ML implementations through hyperparameter tuning, except for SVM with weighted aggregation. Meta-level and statistical features barely show any statistically significant impact on the accuracy of the two tree-based ML algorithms (Decision tree and Random forest), except for implementation through hyperparameter tuning for the weighted aggregation. When we excluded some datasets based on the imbalanced statistics and a significantly higher contribution of one attribute compared to others to the classification performance, we found a significant effect of the meta-level ratio feature and statistical mean and standard deviation features on SVM, LR and KNN accuracy in many cases. Our findings open a new research direction in understanding how dataset characteristics affect ML performance and will help researchers select appropriate ML algorithms for a possible optimal accuracy outcome. [ABSTRACT FROM AUTHOR]

The demand for high-performance displays is driving the development of glass substrates with extremely low strain variation during manufacturing. To achieve the necessary balance of physical properties to resist strain, the traditional compositional design methods for glass substrates like trial-and-error and classical computational techniques should be optimized. As an alternative approach, machine learning (ML) algorithms have emerged for designing new glass compositions. In this work, we conduct ML research focused on the compositional design for high-performance glass substrates. We employ three ML algorithms i.e., Random Forest (RF), Classification And Regression Tree (CART) and k-Nearest Neighbors (k-NN) to predict five physical properties that are key to the performance of alkaline-free aluminosilicate glass substrates. By using only small-to-medium dataset sizes, our model reaches a high coefficient of determination of 0.9879. Furthermore, our model achieves an accurate generalized prediction of 50 experimental data and enables the prediction and design of glass substrate compositions with advanced comprehensive properties. [ABSTRACT FROM AUTHOR]

The north coast of Java is the center of economic activity in Indonesia. This area is dynamic and sensitive to various geo-bio-physical aspects. Therefore, a vulnerability study in this area is necessary. This study proposes a machine learning tree-based algorithms modeling approach for Coastal Vulnerability Assessment (CVA) and mapping. The tree-based algorithms used are Gradient Tree Boost (GTB), Classification and Regression Trees (CART), and Random Forest (RF). The study utilized the Google Earth Engine (GEE) platform and twelve variables as input. The prediction results of each of these modeling algorithms have been compared and evaluated to determine the most optimal performance and accuracy. Reference data was obtained from the Ministry of Maritime Affairs and Fisheries of the Republic of Indonesia (KKP). Approximately 70% of the reference data was allocated for training, while the remaining 30% was designated for validation. The CVA assessment yielded overall accuracies of 80.22%, 77.40%, and 71.18% based on the RF, GTB, and CART algorithms, respectively. Meanwhile, the Kappa Index for these three algorithms was 0.72, 0.67, and 0.58, indicating that the models have adequately classified the data. The research outcomes are anticipated to offer insights into the potential utilization of machine learning technology for vulnerability assessment and mapping, contributing to the management of coastal environmental issues. [ABSTRACT FROM AUTHOR]

To meet the rising power demand, it has become mandatory to consider alternative solutions to the existing transmission line framework. Six phase transmission line can be a potential solution due to its ability to transfer 73% more power relative to classical three phase double circuit system. However, due to the increased number of series and shunt faults in the six-phase transmission line, the development of an accurate protection scheme is quite challenging in a six-phase system. Also, because of the larger number of phases, the possibility of non-typical shunt faults, viz. evolving and cross-country faults, are quite high in a six-phase system. Hence, the significance of developing a protection scheme with increased sensitivity against all types of faults (series, shunts, evolving, and cross-country) has become vital for the acceptance of six-phase transmission lines by power utilities. In this regard, a random forest (RF)-based classifier and regression approach have been respectively proposed for the detection/classification and location estimation of all possible faults in a six-phase transmission line. Separate RF classifier modules have been developed for shunt and series faults for detection and classification, followed by further categorization of shunt faults into evolving and cross-country faults. For location estimation, the high dimensional nonlinear mapping characteristics of the Regression Tree (RT) have been utilized. The use of single-end data, avoids the issue of latency and data loss associated with communication assisted protection schemes. The proposed RF-based scheme's efficacy has been tested for various types of typical and non-typical shunt and series faults, with a wide range of fault and power system parameters. The test findings of the RF-based protection illustrate that the protection scheme is resistant to a wide variation in the fault types, fault conditions, extent of nonlinear loading, and noise intrusion. [ABSTRACT FROM AUTHOR]

Healthcare demand is growing in Australia and across the world. In Australia, the healthcare system comprises a mix of private and public organizations, such as hospitals, clinics, and aged care facilities. The Australian healthcare system is quite affordable and accessible because a large proportion of the expenditure, around 68%, is funded by the Australian government. The healthcare expenditure in 2015-16 was AUD 170.4 billion which was 10.0% of the GDP. Soaring healthcare costs and growing demand for services are increasing the pressure on the sustainability of the government-funded healthcare system. To be sustainable, we need to be more efficient in delivering healthcare services. We can schedule the care delivery process optimally and subsequently improve the efficiency of the system if demand for services is well known. However, there is a randomness in demand for services, and it is a cause of inefficiency in the healthcare delivery process. Soaring healthcare costs and the growing demand for services require us to use healthcare resources more efficiently. Randomness in resource requirements makes the care delivery process less efficient. Our aim is to reduce the uncertainty in patients’ resource requirements, and we achieve that objective by classifying patients into similar resource user groups. The conventional random forest., k-nearest neighbourhood (KNN) methods were resulted in poor classification, prediction performance. In this work, we develop a two-stage classification model to classify patients into lower variability resource user groups by using electronic patient record. There are various statistical tools for classifying patients into lower variability resource user groups. However, classification and regression tree (CART) analysis is a more suitable method for analyzing healthcare data because it has some distinct features. For example, it can handle the interaction between predictor variables naturally, it is nonparametric in nature, and it is relatively insensitive to the curse of dimensionality. [ABSTRACT FROM AUTHOR]

• The potential of optical and SAR data was explored for estimating grassland height. • The RF, GBRT, and CART models were used to estimate grassland height. • GBRT model performed than RF and CART models for grassland height estimation. • The optical remote sensing data performed better than SAR remote sensing data. • Coupling optical and SAR data can effectively improve the estimation accuracy. Grassland height is an important indicator used to evaluate ecological environments in grasslands. The study explored the potential of optical and SAR (synthetic aperture radar) remote sensing data in estimating grassland height with machine learning methods and constructed the grassland height remote sensing inversion model. Then the mean grassland height in August 2015 was estimated with the inversion model in Inner Mongolia Autonomous Region, China. Among classification and regression tree (CART), random forest (RF), and gradient boosting regression tree (GBRT) models, the GBRT model had the highest estimation accuracy. There was little correlation between SAR data and grassland height, and SAR data produced poor accuracy for grassland height estimation. The grassland height could be better estimated by optical remote sensing data. The best estimation accuracy (GBRT model: for training data: R2 = 0.71, RMSE = 3.58 cm, P < 0.01; for test data: R2 = 0.58, RMSE = 3.94 cm, P < 0.01) was achieved by the combination of optical and SAR remote sensing data. However, SAR data played an auxiliary role, and the estimation of grassland height was mainly realized by optical data. The average vegetation height of the grasslands in Inner Mongolia in August 2015 was 19.15 cm, gradually decreasing from northeast to southwest. The study proposed a high-precision method for estimating grassland height, which provided a basis for studying grassland environments and conditions. [ABSTRACT FROM AUTHOR]

Background: Poverty contributes to the transmission of schistosomiasis via multiple pathways, with the insufficiency of appropriate interventions being a crucial factor. The aim of this article is to provide more economical and feasible intervention measures for endemic areas with varying levels of poverty. Methods: We collected and analyzed the prevalence patterns along with the cost of control measures in 11 counties over the last 20 years in China. Seven machine learning models, including XGBoost, support vector machine, generalized linear model, regression tree, random forest, gradient boosting machine and neural network, were used for developing model and calculate marginal benefits. Results: The XGBoost model had the highest prediction accuracy with an R2 of 0.7308. Results showed that risk surveillance, snail control with molluscicides and treatment were the most effective interventions in controlling schistosomiasis prevalence. The best combination of interventions was interlacing seven interventions, including risk surveillance, treatment, toilet construction, health education, snail control with molluscicides, cattle slaughter and animal chemotherapy. The marginal benefit of risk surveillance is the most effective intervention among nine interventions, which was influenced by the prevalence of schistosomiasis and cost. Conclusions: In the elimination phase of the national schistosomiasis program, emphasizing risk surveillance holds significant importance in terms of cost-saving. [ABSTRACT FROM AUTHOR]

Due to the transition toward the Internet of Everything (IOE), the prediction of energy consumed by household appliances has become a progressively more difficult topic to model. Even with advancements in data analytics and machine learning, several challenges remain to be addressed. Therefore, providing highly accurate and optimized models has become the primary research goal of many studies. This paper analyzes appliance energy consumption through a variety of machine learning-based strategies. Utilizing data recorded from a single-family home, input variables comprised internal temperatures and humidities, lighting consumption, and outdoor conditions including wind speed, visibility, and pressure. Various models were trained and evaluated: (a) multiple linear regression, (b) support vector regression, (c) random forest, (d) gradient boosting, (e) xgboost, and (f) the extra trees regressor. Both feature engineering and hyperparameter tuning methodologies were applied to not only extend existing features but also create new ones that provided improved model performance across all metrics: root mean square error (RMSE), coefficient of determination (R2), mean absolute error (MAE), and mean absolute percentage error (MAPE). The best model (extra trees) was able to explain 99% of the variance in the training set and 66% in the testing set when using all the predictors. The results were compared with those obtained using a similar methodology. The objective of performing these actions was to show a unique perspective in simulating building performance through data-driven models, identifying how to maximize predictive performance through the use of machine learning-based strategies, as well as understanding the potential benefits of utilizing different models. [ABSTRACT FROM AUTHOR]

Agricultural and environmental literacy are essential public goods, but associated education efforts struggle to reach broad audiences. Understanding learner backgrounds and lived experiences can help address this challenge. We assessed the relative importance of demographics, parent views of agriculture, interactions with farmers and parents, and learning setting in predicting agricultural literacy among 525 elementary school children in North Carolina, USA. We used classification and regression trees and random forest models, which account for non-linear and interacting relationships. Knowing a farmer and engagement with parents were more predictive of children agricultural literacy than demographics, countering historically held deficit-based assumptions around agricultural and environmental literacy. [ABSTRACT FROM AUTHOR]

Many species rely on acoustic communication to coordinate activities and communicate to conspecifics. Cataloging vocal behavior is a first step towards understanding how individuals communicate information and how communication may be degraded by anthropogenic noise. The Cook Inlet beluga population is endangered with an estimated 331 individuals. Anthropogenic noise is considered a threat for this population and can negatively impact communication. To characterize this population's vocal behavior, vocalizations were measured and classified into three categories: whistles (n = 1264, 77%), pulsed calls (n = 354, 22%), and combined calls (n = 15, 1%), resulting in 41 call types. Two quantitative analyses were conducted to compare with the manual classification. A classification and regression tree and Random Forest had a 95% and 85% agreement with the manual classification, respectively. The most common call types per category were then used to investigate masking by commercial ship noise. Results indicate that these call types were partially masked by distant ship noise and completely masked by close ship noise in the frequency range of 0–12 kHz. Understanding vocal behavior and the effects of masking in Cook Inlet belugas provides important information supporting the management of this endangered population. [ABSTRACT FROM AUTHOR]

Wildfire has extensive and profound impacts on forest structure and function. Therefore, it is important to study the spatial and temporal patterns of forest fire regimes and their drivers in order to better understand the feedbacks between climate change, fire disturbance, and forest ecosystems. Based on the Global Fire Atlas dataset, three forest fire regime components (fire occurrence density, burned rate, and median fire size) were extracted for China from 2003 to 2016. Three statistical models (Boosted Regression Tree, Random Forest, and Support Vector Machine) were used to systematically analyze the relationships between patterns of forest fire disturbance and climate, human activities, vegetation, and topography in China, as well as their spatial heterogeneity in different climatic regions. The results indicate that the spatial distribution of forest fires is heterogeneous, and different forest fire regime components are predicted by different factors. At the national level, the distribution of forest fire regimes mainly corresponds to climatic factors, although the relationship between median fire size and predictors is obscure. At the scale of each ecoregion, the main climate predictors of forest fire occurrence density and burned rate change from temperature in the north to temperature and precipitation in the south. Median fire size varies with elevation and temperature in the south. These results demonstrate that the spatial heterogeneity of predictors and scaling effects must be fully considered in the study of forest fire disturbance. [ABSTRACT FROM AUTHOR]

The Earth‐abundant element‐based Cu2ZnSn(S,Se)4 (CZTSSe) absorber is considered as a promising material for thin‐film solar cells (TFSCs). The current record power conversion efficiency (PCE) of CZTSSe TFSCs is ≈13%, and it's still lower than CdTe and CIGS‐based TFSCs. A further breakthrough in its PCE mainly relies on deep insights into the various device fabrication conditions; accordingly, the experimental–oriented machine learning (ML) approach can be an effective way to discover key governing factors in improving PCE. The present work aims to identify the key governing factors throughout the device fabrication processes and apply them to break the saturated PCE for CZTSSe TFSCs. For realization, over 25,000 data points were broadly collected by fabricating more than 1300 CZTSSe TFSC devices and analyzed them using various ML techniques. Through extensive ML analysis, the i‐ZnO thickness is found to be the first, while Zn/Sn compositional ratio and sulfo‐selenization temperature are other key governing factors under thin or thick i‐ZnO thickness to achieve over 11% PCE. Based on these key governing factors, the applied random forest ML prediction model for PCE showed Adj. R2 = >0.96. Finally, the best‐predicted ML conditions considered for experimental validation showed well‐matched experimental outcomes with different ML models. [ABSTRACT FROM AUTHOR]

The study aimed to identify critical factors associated with the surgical stability of pogonion (Pog) by applying machine learning (ML) to predict relapse following two-jaw orthognathic surgery (2 J-OGJ). The sample set comprised 227 patients (110 males and 117 females, 207 training and 20 test sets). Using lateral cephalograms taken at the initial evaluation (T0), pretreatment (T1), after (T2) 2 J-OGS, and post treatment (T3), 55 linear and angular skeletal and dental surgical movements (T2-T1) were measured. Six ML modes were utilized, including classification and regression trees (CART), conditional inference tree (CTREE), and random forest (RF). The training samples were classified into three groups; highly significant (HS) (≥ 4), significant (S) (≥ 2 and < 4), and insignificant (N), depending on Pog relapse. RF indicated that the most important variable that affected relapse rank prediction was ramus inclination (RI), CTREE and CART revealed that a clockwise rotation of more than 3.7 and 1.8 degrees of RI was a risk factor for HS and S groups, respectively. RF, CTREE, and CART were practical tools for predicting surgical stability. More than 1.8 degrees of CW rotation of the ramus during surgery would lead to significant Pog relapse. [ABSTRACT FROM AUTHOR]

Tree ensembles can be interpreted as implicit kernel generators, where the ensuing proximity matrix represents the data-driven tree ensemble kernel. Focus of our work is the utility of tree based ensembles as kernel generators that (in conjunction with a regularized linear model) enable kernel learning. We elucidate the performance of the tree based random forest (RF) and gradient boosted tree (GBT) kernels in a comprehensive simulation study comprising of continuous and binary targets. We show that for continuous targets (regression), this kernel learning approach is competitive to the respective tree ensemble in higher dimensional scenarios, particularly in cases with larger number of noisy features. For the binary target (classification), the tree ensemble based kernels and their respective ensembles exhibit comparable performance.We provide the results fromseveral real life datasets for regression and classification relevant for biopharmaceutical and biomedical applications, that are in line with the simulations to show how these insights may be leveraged in practice. We discuss general applicability and extensions of the tree ensemble based kernels for survival targets and interpretable landmarking in classification and regression. Finally, we outline future research for kernel learning due to feature space partitionings. [ABSTRACT FROM AUTHOR]

Among the critical global crises curbing world development and sustainability, air quality degradation has been a long-lasting and increasingly urgent one and it has been sufficiently proven to pose severe threats to human health and social welfare. A higher level of model prediction accuracy can play a fundamental role in air quality assessment and enhancing human well-being. In this paper, four types of machine learning models—random forest model, ridge regression model, support vector machine model, extremely randomized trees model—were adopted to predict PM2.5 concentration in ten cities in the Jing-Jin-Ji region of north China based on multi-sources spatiotemporal data including air quality and meteorological data in time series. Data were fed into the model by using the rolling prediction method which is proven to improve prediction accuracy in our experiments. Lastly, the comparative experiments show that at the city level, RF and ExtraTrees models have better predictive results with lower mean absolute error (MAE), root mean square error (RMSE), and higher index of agreement (IA) compared to other selected models. For seasonality, level four models all have the best prediction performances in winter time and the worst in summer time, and RF models have the best prediction performance with the IA ranging from 0.93 to 0.98 with an MAE of 5.91 to 11.68 μg/m3. Consequently, the demonstration of how each model performs differently in each city and each season is expected to shed light on environmental policy implications. [ABSTRACT FROM AUTHOR]

Nearly 60% of commercial customers are connected to a low-voltage network in Poland with a contractual capacity of more than 40 kW and are assigned a fixed tariff with flat prices for the whole year, no matter the usage volume. With smart meters, more data about how businesses use energy are becoming available to both energy providers and customers. This enables innovation in the structure and type of tariffs on offer in the energy market. Customers can explore their usage patterns to choose the most suitable tariff to benefit from lower prices and thus generate savings. In this paper, we analyzed whether customers' electricity usage matched their optimal tariff and further investigated which of them could benefit or lose from switching the tariff based on the real dataset with the hourly energy readings of 1212 commercial entities in Poland recorded between 2016 and 2019. Three modelling approaches, i.e., the k-nearest neighbors, classification tree and random forest, were tested for optimal tariff classification, while for the benchmark, we used a simple approach, in which the tariff was proposed based on the customers' previous electricity usage. The main findings from the research are threefold: (1) out of all the analyzed entities, on average, 76% of them could have benefited from the tariff switching, which suggests that customers may not be aware of the tariff change benefits, or they had chosen a tariff plan that was not tailored to them; (2) a random forest model offers a viable approach to accurate tariff classification; (3) the policy implication from the research is the need to increase the customers' awareness about the tariffs and propose reliable tools for selecting the optimal tariff. [ABSTRACT FROM AUTHOR]

In recent years precise and up-to-date information regarding seabed depth has become more and more important for companies and institutions that operate on coastlines. While direct, in-situ measurements are performed regularly, they are expensive, time-consuming and impractical to be performed in short time intervals. At the same time, an ever-increasing amount of satellite imaging data becomes available. With these images, it became possible to develop bathymetry estimation algorithms that can predict seabed depth and utilize them systematically. Since there are a number of theoretical approaches, physical models, and empirical techniques to use satellite observations in order to estimate depth in the coastal zone, the presented article compares the performance and precision of the most common one to modern machine learning algorithms. More specifically, the models based on shallow neural networks, decision trees and Random Forest algorithms have been proposed, investigated and confronted with the performance of pure analytical models. The particular proposed machine learning models differ also in a set of satellite data bands used as an input as well as in applying or not geographical weighting in the learning process. The obtained results point towards the best performance of the regression tree algorithm that incorporated as inputs information about data localization, raw reflectance data from four satellite data bands and a quotient of logarithms of B2 and B3 bands. The study for the paper was performed in relatively optically difficult and spatially variant conditions of the south Baltic coastline starting at Szczecin, Poland on the west (53°26'17" N, 14°32'32" E) to Hel peninsula (54°43'04,3774" N 18°37'56,9175" E). The reference bathymetry data was acquired from Polish Marine Administration. It was obtained through profile probing with single-beam sonar or direct in-situ probing. [ABSTRACT FROM AUTHOR]

We present a Bayesian nonparametric model for conditional distribution estimation using Bayesian additive regression trees (BART). The generative model we use is based on rejection sampling from a base model. Like other BART models, our model is flexible, has a default prior specification, and is computationally convenient. To address the distinguished role of the response in our BART model, we introduce an approach to targeted smoothing of BART models which is of independent interest. We study the proposed model theoretically and provide sufficient conditions for the posterior distribution to concentrate at close to the minimax optimal rate adaptively over smoothness classes in the high-dimensional regime in which many predictors are irrelevant. To fit our model, we propose a data augmentation algorithm which allows for existing BART samplers to be extended with minimal effort. We illustrate the performance of our methodology on simulated data and use it to study the relationship between education and body mass index using data from the medical expenditure panel survey (MEPS). for this article are available online. [ABSTRACT FROM AUTHOR]

Gradient tree boosting is a powerful machine learning technique that has shown good performance in predicting a variety of outcomes. However, when applied to hierarchical (e.g., longitudinal or clustered) data, the predictive performance of gradient tree boosting may be harmed by ignoring the hierarchical structure, and may be improved by accounting for it. Tree-based methods such as regression trees and random forests have already been extended to hierarchical data settings by combining them with the linear mixed effects model (MEM). In the present article, we add to this literature by proposing two algorithms to estimate a combination of the MEM and gradient tree boosting. We report on two simulation studies that (i) investigate the predictive performance of the two MEM boosting algorithms and (ii) compare them to standard gradient tree boosting, standard random forest, and other existing methods for hierarchical data (MEM, MEM random forests, model-based boosting, Bayesian additive regression trees [BART]). We found substantial improvements in the predictive performance of our MEM boosting algorithms over standard boosting when the random effects were non-negligible. MEM boosting as well as BART showed a predictive performance similar to the correctly specified MEM (i.e., the benchmark model), and overall outperformed the model-based boosting and random forest approaches. [ABSTRACT FROM AUTHOR]

The article "Drought and life-history strategies in Heliophila (Brassicaceae)" by Ladislav Mucina et al. critiques a study by Monroe et al. that claims to predict life-history strategies in plants based on drought regimes. Mucina et al. argue that Monroe et al.'s study has methodological and conceptual errors, including inaccurate distribution data, uncritical application of phylogenetic data, and the use of short-term climatic data to infer evolutionary change. They also question the use of the Vegetative Health Index as a proxy for drought stress. Mucina et al. re-analyze the relationship between climatic and nonclimatic variables and find that long-term rainfall during specific months is a more effective variable in driving the separation between annual and perennial species. They conclude that drought frequency is not a reliable factor in explaining plant life histories. [Extracted from the article]

Atrial fibrillation (AF) is a life threatening disease and can cause stroke, heart failure, and sometimes death. To reduce the rate of mortality and morbidity due to increased prevalence of AF, early detection of the same becomes a prior concern. Traditional machine learning (TML) algorithms and ensemble machine learning (EML) algorithms are proposed to detect AF in this article. The performances of both these methods are compared in this study. Methodology involves computation of RR interval features extracted from electrocardiogram and its classification into: normal, AF, and other rhythms. TML techniques such as Classification and Regression Tree, K Nearest Neighbor, C4.5, Iterative Dichotomiser 3, Support Vector Machine and EML classifier such as Random Forest (RF), and Rotation Forest are used for classification. The proposed method is evaluated using PhysioNet challenge 2017. During the tenfold cross validation, it is observed that RF classifier provided good classification accuracy of 99.10% with area under the curve of 0.998. Apart from contributing a new methodology, the proposed study also experimentally proves higher performance with ensemble learning method, RF. The methodology has many applications in health care management systems including defibrillators, cardiac pacemakers, etc. [ABSTRACT FROM AUTHOR]

Glacier mass balance is typically estimated using a range of in situ measurements, remote sensing measurements, and physical and temperature index modelling techniques. With improved data collection and access to large datasets, data-driven techniques have recently gained prominence in modelling natural processes. The most common data-driven techniques used today are linear regression models and, to some extent, non-linear machine learning models such as artificial neural networks. However, the entire host of capabilities of machine learning modelling has not been applied to glacier mass balance modelling. This study used monthly meteorological data from ERA5-Land to drive four machine learning models: random forest (ensemble tree type), gradient-boosted regressor (ensemble tree type), support vector machine (kernel type), and artificial neural networks (neural type). We also use ordinary least squares linear regression as a baseline model against which to compare the performance of the machine learning models. Further, we assess the requirement of data for each of the models and the requirement for hyperparameter tuning. Finally, the importance of each meteorological variable in the mass balance estimation for each of the models is estimated using permutation importance. All machine learning models outperform the linear regression model. The neural network model depicted a low bias, suggesting the possibility of enhanced results in the event of biased input data. However, the ensemble tree-based models, random forest and gradient-boosted regressor, outperformed all other models in terms of the evaluation metrics and interpretability of the meteorological variables. The gradient-boosted regression model depicted the best coefficient of determination value of 0.713 and a root mean squared error of 1.071 m w.e. The feature importance values associated with all machine learning models suggested a high importance of meteorological variables associated with ablation. This is in line with predominantly negative mass balance observations. We conclude that machine learning techniques are promising in estimating glacier mass balance and can incorporate information from more significant meteorological variables as opposed to a simplified set of variables used in temperature index models. [ABSTRACT FROM AUTHOR]