Random Forest Machine Learning Algorithm Research Articles

Investigating the Relationship between Urbanization and Air Pollution Using Google Earth Engine Platform: A Case Study of Istanbul

Rapid population growth in megacities such as Istanbul has led to various effects, such as industrialization, urbanization, loss of green areas, increasing vehicle traffic, and higher consumption of fossil fuels. These reasons, along with many other environmental factors, contribute to the rise of air pollution in urban life. This study aimed to examine the relationship between urbanization and air pollution in Istanbul. For this purpose, land cover maps covering Istanbul province were produced using Landsat-5 (TM), Landsat-8 (OLI), and Sentinel-2 (MSI) images for the years 1996 to 2021 at three-year intervals on the Google Earth Engine platform. Land cover for classification purposes was divided into five different classes: forest, water surface, urban area, and bare land, and classified using a random forest machine learning algorithm. To examine the impact of this urban area growth on air pollution, in the second step of the study, the column number density values of Sentinel 5P (TROPOMI) data for SO2, NO2, CO, and O3 gases for 2019, 2020, and 2021 were analyzed. The averages of the data from 39 air pollution monitoring stations across Istanbul were also examined. According to this classification, the urban area expanded from 491 km2 in 1996 to 1222 km2 by 2021. Considering the total surface area of Istanbul province, the urban area, which was 9% in 1996, reached 23% by 2021. The TROPOMI values were calculated as follows: the average column number density values for SO2, NO2, CO, and O3 were 0.0003538 mol/m², 0.0339514 mol/m², 0.0000984 mol/m², and 0.1453515 mol/m², respectively. Similarly, the gas concentrations of SO2, NO2, CO, and O3 measured from the ground stations were calculated as 6.603 µ/m3, 786,815 µ/m3, 43.763 µ/m3 and 45.773 µ/m3, respectively. Correlations between urbanization and TROPOMI values revealed a positive correlation of 0.39, 0.02, and 0.80 for SO2, NO2, and CO gases, while a negative correlation of 0.25 was found for O3 gas. The study also examined correlations between TROPOMI and ground station measurements, resulting in positive correlations of 0.55, 0.66, and 0.16 for SO2, NO2, and CO gases, respectively, while a negative correlation of 0.05 was found for O3 gas. Based on these findings, among the air pollutants studied both through TROPOMI and ground station data, the highest correlation was observed for CO gas.

Shippers/freight forwarders’ acceptance of dedicated rail freight corridors for freight mobility in India

This paper investigates the acceptance of a mega rail freight infrastructure as a sustainable alternative to road transport for domestic freight movements in India. The dedicated rail freight corridors (DFCs) are the freight-only rail corridors proposed by the Indian government to improve freight mobility from a sustainable outlook. A shipper/freight forwarder survey was conducted to gather information on mode attributes and their stated preferences toward DFCs. We employ discrete choice (binary logit) and machine learning algorithms (random forest and extreme gradient boosting) to analyse the choice behaviour. The machine learning methods exhibited higher prediction accuracy, while discrete choice models offered better interpretability. On-time performance and transport costs are crucial factors that influence mode choice. Large-scale companies are more willing to shift to DFCs compared to small and medium firms. The policy scenario analysis indicates that providing a better on-time performance can gain a substantial share of DFCs.

Classification of Agricultural Crops with Random Forest and Support Vector Machine Algorithms Using Sentinel-2 and Landsat-8 Images

Monitoring crop development and mapping cultivated areas are important for reducing risks to food security due to climate change. Remote sensing techniques contribute significantly to the efficient and effective management of agricultural production. In this study, agricultural fields (sunflower, wheat, maize, oat, chickpea, sugar beet, alfalfa, onion, fallow) and other fields (non-agricultural, pasture, lake) were identified by using Random Forest (RF) and Support Vector Machines (SVM) machine learning algorithms with Sentinel-2 and Landsat-8 images in the area covering Polatlı, Haymana and Gölbaşı districts of Ankara province Multi-temporal images were used to distinguish winter and summer crops, taking into account crop development periods. As a result of classification; the overall accuracy of RF and SVM models with S2 images are 89.5% and 84.6% and kappa coefficients are 0.88 and 0.83, while the overall accuracy of RF and SVM models with L8 images are 79% and 78.1% and kappa coefficients are 0.76 and 0.75. RF model was found to have higher prediction accuracy than SVM. Sentinel-2 imagery has a higher accuracy in all classes compared to Landsat-8, indicating that Sentinel-2 imagery with its high temporal and spatial resolution is more suitable and has a great potential for agricultural crop pattern detection.

Enhancing PVT property predictions for black oil reservoirs through the application of supervised machine learning techniques

The study presents an innovative approach utilizing machine learning (ML) techniques to forecast essential characteristics of black oil reservoirs, namely the oil formation volume factor (Bo) and the solution gas-oil ratio (Rs). A suite of four established ML algorithms-Random Forest, Linear Regression, Ada Boost, and Gradient Boost were extensively conditioned using a dataset that includes 297 individual data points and evaluation is done using test data and a blind test of models. The study showcases the practical efficacy of ML in accurately predicting Bo and Rs, even with a relatively modest dataset. The potential of ML models to save time and costs by reducing the dependence on laborious laboratory measurements is underscored, presenting a transformative shift in reservoir engineering practices. Furthermore, the current research extends its impact by comparing ML predictions with various empirical correlations and actual values, demonstrating that ML models, particularly Gradient Boost, consistently outperform traditional correlations and it can be used to predict values in the undersaturated region too. It is observed from the blind test results that for the prediction of Bo Gradient boost and random forest models (R2 0.97, RMSE 0.01) is a preferable choice and for the prediction of Rs the Gradient boost model (R2 0.97, RMSE 22) can be preferred. The adaptability of ML models to integrate new data over time is a significant feature, ensuring continuous improvement in accuracy. This study adds to the expanding reservoir engineering knowledge base, highlighting the practical use of machine learning for accurately predicting crucial reservoir properties. Its significance goes beyond theoretical progress, providing tangible tools for reservoir engineers and geologists to improve decision-making in the management of oil and gas reservoirs.

Proteomics profiling and machine learning in nusinersen-treated patients with spinal muscular atrophy

AimThe availability of disease-modifying therapies and newborn screening programs for spinal muscular atrophy (SMA) has generated an urgent need for reliable prognostic biomarkers to classify patients according to disease severity. We aim to identify cerebrospinal fluid (CSF) prognostic protein biomarkers in CSF samples of SMA patients collected at baseline (T0), and to describe proteomic profile changes and biological pathways influenced by nusinersen before the sixth nusinersen infusion (T302).MethodsIn this multicenter retrospective longitudinal study, we employed an untargeted liquid chromatography mass spectrometry (LC-MS)-based proteomic approach on CSF samples collected from 61 SMA patients treated with nusinersen (SMA1 n=19, SMA2 n=19, SMA3 n=23) at T0 at T302. The Random Forest (RF) machine learning algorithm and pathway enrichment analysis were applied for analysis.ResultsThe RF algorithm, applied to the protein expression profile of naïve patients, revealed several proteins that could classify the different types of SMA according to their differential abundance at T0. Analysis of changes in proteomic profiles identified a total of 147 differentially expressed proteins after nusinersen treatment in SMA1, 135 in SMA2, and 289 in SMA3.Overall, nusinersen-induced changes on proteomic profile were consistent with i) common effects observed in allSMA types (i.e. regulation of axonogenesis), and ii) disease severity-specific changes, namely regulation of glucose metabolism in SMA1, of coagulation processes in SMA2, and of complement cascade in SMA3.ConclusionsThis untargeted LC-MS proteomic profiling in the CSF of SMA patients revealed differences in protein expression in naïve patients and showed nusinersen-related modulation in several biological processes after 10 months of treatment. Further confirmatory studies are needed to validate these results in larger number of patients and over abroader timeframe.

Wind speed prediction in some major cities in Africa using Linear Regression and Random Forest algorithms

Globally, wind energy if properly harnessed, could serve as a source of energy generation in Africa. This study compared the performance of two Machine Learning (ML) algorithms (Linear regression and Random Forest) in predicting wind speed in five major cities in Africa (Yaoundé, Pretoria, Nairobi, Cairo and Abuja). Wind data were collected between January 1, 2000, and December 31, 2022, using the Solar Radiation Data Archive. The data preprocessing was carried out with 80% of the data used for training and 20% for validation. The performance of these ML algorithms was evaluated using Mean Square Error (MSE), Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE) and coefficient of determination (R2). The result shows that Nairobi (3.814795 m/s) closely followed by Cairo (3.606453 m/s) has the highest mean wind speed while Yaoundé (1.090512 m/s) has the lowest. Based on the performance metrics used, the two Machine Learning algorithms were competitive. Still, the Linear Regression (LR) algorithm outperformed the Random Forest Algorithm in predicting wind speed in all the selected major African cities. In Yaoundé (RMSE = 0.3892, MAE= 0.3001, MAPE =0.5030), Pretoria (RMSE=1.2339, MAE=0.9480, MAPE=0.7450) Nairobi (RMSE= 0.4223, MAE =0.6499, MAPE =0.1872), Nairobi (RMSE=0.6499, MAE=0.5171, MAPE =0.1872), Cairo (RMSE =1.0909, MAE =0.8544, MAPE =0.3541) and Abuja (RMSE = 0.70245, MAE =0.5441, MAPE= 0.4515) the Linear regression algorithms was found to outperformed Random Forest Regression. Therefore, the Linear regression algorithm is more reliable in predicting wind speed compared with the Random Forest regression.

Early and swift identification of fungal-infection using infrared spectroscopy

Fungal pathogens pose significant threats to agricultural crops and food products, leading to economic losses, compromised food quality, and health hazards. Early detection is crucial for effective control and treatment. This study explores Fourier transform infrared-attenuated total reflectance (FTIR-ATR) spectroscopy for rapid fungal detection in bread. Using a machine learning algorithm (Random Forest), FTIR-ATR accurately distinguished between pure and infected bread samples, achieving 86% overall accuracy and 84% accuracy in identifying specific fungi like Rhizopus and Aspergillus on the first day of infection. These findings highlight FTIR-ATR’s potential for early fungal infection detection, promising improved food quality and reduced economic losses through timely intervention.

A holistic approach to software fault prediction with dynamic classification

Software Fault Prediction is a critical domain in machine learning aimed at pre-emptively identifying and mitigating software faults. This study addresses challenges related to imbalanced datasets and feature selection, significantly enhancing the effectiveness of fault prediction models. We mitigate class imbalance in the Unified Dataset using the Random-Over Sampling technique, resulting in superior accuracy for minority-class predictions. Additionally, we employ the innovative Ant-Colony Optimization algorithm (ACO) for feature selection, extracting pertinent features to amplify model performance. Recognizing the limitations of individual machine learning models, we introduce the Dynamic Classifier, a ground-breaking ensemble that combines predictions from multiple algorithms, elevating fault prediction precision. Model parameters are fine-tuned using the Grid-Search Method, achieving an accuracy of 94.129% and superior overall performance compared to random forest, decision tree and other standard machine learning algorithms. The core contribution of this study lies in the comparative analysis, pitting our Dynamic Classifier against Standard Algorithms using diverse performance metrics. The results unequivocally establish the Dynamic Classifier as a frontrunner, highlighting its prowess in fault prediction. In conclusion, this research introduces a comprehensive and innovative approach to software fault prediction. It pioneers the resolution of class imbalance, employs cutting-edge feature selection, and introduces dynamic ensemble classifiers. The proposed methodology, showcasing a significant advancement in performance over existing methods, illuminates the path toward developing more accurate and efficient fault prediction models.

SCARF: A new algorithm for continuous prediction of biomass dynamics using machine learning and Landsat time series

We developed the SCARF (Spatial Mismatch and Systematic Prediction Error Corrected cAscade Random Forests) algorithm for continuous prediction of biomass dynamics using machine learning and Landsat Time Series (LTS). Our approach addresses the challenges posed by the cloudy subtropical forests in southern China, where monitoring biomass dynamics is notoriously difficult. To derive spectral-temporal features from the LTS, we applied the Continuous Change Detection and Classification (CCDC) algorithm (Zhu and Woodcock, 2014). Subsequently, we employed the cascade random forests machine learning algorithm for biomass prediction. This new approach corrects the spatial mismatch effects between plots and Landsat pixels as well as the systematic prediction errors in the machine learning model. As a result, it substantially enhances biomass prediction accuracy, with a coefficient of determination (R2) of 0.83 and a root mean square error (RMSE) of 6.27 Mg ha-1. In comparison, the commonly used random forests approach yields an R2 of 0.47 and RMSE of 8.52 Mg ha-1. Additionally, it provides reliable spatial prediction beyond the model-training area, achieving an R2 of 0.79 and an RMSE of 6.62 Mg ha-1. Furthermore, we demonstrate that modeling five different forest age groups separately further improves prediction accuracies, resulting in an increased R2 of 0.87 and a reduced RMSE of 3.65 Mg ha-1. A comparison of the allometric model prediction from the field plots and those from the SCARF model revealed a strong agreement, indicating that this approach can provide a temporally continuous prediction of biomass dynamics. Our study presents a robust method for continuous, reliable, and explicit spatiotemporal prediction of biomass dynamics in cloudy subtropical forests using LTS.

Mussel culture monitoring with semi-supervised machine learning on multibeam echosounder data using label spreading

High diversity seabed habitats, such as shellfish aggregations, play a significant role in marine ecosystem sustainability but are susceptible to bottom disturbance induced by anthropogenic activities. Regular monitoring of these habitats with effective mapping methods is therefore essential. Multibeam echosounder (MBES) has been widely used in recent decades for seabed characterization due to its non-destructive manner and extensive spatial coverage compared to traditional methods like bottom sampling. Nevertheless, bottom sampling remains essential to link ground truth with acoustic seabed classification. Using seabed samples and MBES measurements, machine learning techniques are commonly employed to model their relationships and generate classification maps of an extended seabed. However, limited ground truth data, resulting from constraints in regulations, budget, or time, may impede the development of robust machine learning models. To address this challenge, we applied a semi-supervised machine learning method to classify seabed sediments of a blue mussel (Mytilus edulis) cultivation area in the Oosterschelde, the Netherlands. We utilized nine boxcore samples to generate pseudo-labels on MBES data. These pseudo-labels enlarged the training data size, facilitated the training of three comprehensive machine learning algorithms (Gradient Boosting, Random Forest, and Support Vector Machine), and helped to classify the study site into mussel and non-mussel areas. We found the geomorphological and backscatter-related features to be complementary for mussel culture detection. Our classification results were demonstrated effective through expert knowledge of this cultivation area and brought insights for future research on natural mussel habitats.

Random forest machine learning for maize yield and agronomic efficiency prediction in Ghana

Maize (Zea mays) is an important staple crop for food security in Sub-Saharan Africa. However, there is need to increase production to feed a growing population. In Ghana, this is mainly done by increasing acreage with adverse environmental consequences, rather than yield increment per unit area. Accurate prediction of maize yields and nutrient use efficiency in production is critical to making informed decisions toward economic and ecological sustainability. We trained the random forest machine learning algorithm to predict maize yield and agronomic efficiency in Ghana using soil, climate, environment, and management factors, including fertilizer application. We calibrated and evaluated the performance of the random forest machine learning algorithm using a 5 × 10-fold nested cross-validation approach. Data from 482 maize field trials consisting of 3136 georeferenced treatment plots conducted in Ghana from 1991 to 2020 were used to train the algorithm, identify important predictor variables, and quantify the uncertainties associated with the random forest predictions. The mean error, root mean squared error, model efficiency coefficient and 90 % prediction interval coverage probability were calculated. The results obtained on test data demonstrate good prediction performance for yield (MEC = 0.81) and moderate performance for agronomic efficiency (MEC = 0.63, 0.55 and 0.54 for AE-N, AE-P and AE-K, respectively). We found that climatic variables were less important predictors than soil variables for yield prediction, but temperature was of key importance to yield prediction and rainfall to agronomic efficiency. The developed random forest models provided a better understanding of the drivers of maize yield and agronomic efficiency in a tropical climate and an insight towards improving fertilizer recommendations for sustainable maize production and food security in Sub-Saharan Africa.

Prediction of soil organic matter using Landsat 8 data and machine learning algorithms in typical karst cropland in China

Soil organic matter (SOM) is crucial for karst ecosystems, affecting cropland health, climate change mitigation, and rocky desertification control. However, there are limited research on cropland SOM prediction in karst areas with complex topography and diverse microclimates. Here, we compared the performance of four machine learning algorithms—random forest (RF), support vector regression (SVR), multilayer perceptron regression (MLP), and gradient boosting regression trees (GBRT)—for predicting cropland SOM in a typical karst landform area in 2019. Our results indicated that the GBRT model achieved the highest prediction accuracy with an R² of 0.69, MAE of 2.19 g/kg, RMSE of 3.37 g/kg, and LCCC of 0.82. Using the GBRT model and spatial data on climate, topography, and remote sensing, we predicted SOM for each 30 m × 30 m grid cell. The analysis revealed higher SOM content in the northeastern and southwestern regions and lower content in the central area, ranging from 13.95 to 47.81 g/kg, with an average of 27.16 g/kg. Lime soil had the highest SOM content, while purple soil had the lowest. Paddy fields showed significantly higher SOM than dry land. Over the past 40 years, SOM content has slightly increased, while its spatial distribution has remained stable.

Unveiling Dropout Patterns: A Python-Powered Analysis of Student Attrition in School Education

Student dropout rates pose a significant challenge in Gujarat, India, affecting both societal welfare and economic development. This study aims to diagnose and predict student dropout patterns using data analytics and machine learning (ML) algorithms. Extensive datasets from Gujarat's educational authorities are preprocessed with Python libraries such as Pandas and NumPy, ensuring quality and consistency. Exploratory data analysis (EDA) uncovers insights and trends, while feature engineering extracts relevant characteristics for ML models. The study implements ML algorithms Random Forest (RF), Decision Trees (DT), Logistic Regression (LR), K-Nearest Neighbors (KNN), AdaBoost, XGBoost, Support Vector Machine(SVM) to construct predictive models. The findings reveal critical factors influencing dropout rates and assess model performance using model accuracy. This research provides novel insights into student dropout rates in Gujarat, offering a comprehensive, data-driven approach to address educational attrition and guide targeted intervention strategies and policy recommendations for improving educational outcomes in the region.

Aeolian Desertification Dynamics from 1995 to 2020 in Northern China: Classification Using a Random Forest Machine Learning Algorithm Based on Google Earth Engine

Aeolian Desertification Dynamics from 1995 to 2020 in Northern China: Classification Using a Random Forest Machine Learning Algorithm Based on Google Earth Engine

Lipidomics combined with random forest machine learning algorithms to reveal freshness markers for duck eggs during storage in different rearing systems

The differences in lipids in duck eggs between the 2 rearing systems during storage have not been fully studied. Herein, we propose untargeted lipidomics combined with a random forest (RF) algorithm to identify potential marker lipids based on ultra-performance liquid chromatography‒mass spectrometry (UPLPC-MS/MS). A total of 106 and 16 differential lipids (DL) were screened in egg yolk and white, respectively. In yolk, metabolic pathway analysis of DLs revealed that glycerophospholipid metabolism and sphingolipid metabolism were the key metabolic pathways in the traditional free-range system (TFS) during storage, glycosylphosphatidylinositol-anchored biosynthesis and glyceride metabolism were the key pathways in the floor-rearing system (FRS). In egg white, the key pathway in both systems is the biosynthesis of unsaturated fatty acids. Combined with RF algorithm, 12 marker lipids were screened during storage. Therefore, this study elucidates the changes in lipids in duck eggs during storage in 2 rearing systems and provides new ideas for screening marker lipids during storage. This approach is highly important for evaluating the quality of egg and egg products and provides guidance for duck egg production.

Study of machine learning on the photocatalytic activity of a novel nanozeolite for the application in the Rhodamine B dye degradation

Contamination of wastewater with organic dyes has caused a serious threat to humans and aquatic life due to the hazardous effect of these contaminants. In this context, the present work aims to carry out a Machine Learning (ML) study to evaluate the photocatalytic activity of a nanozeolite (nANA) in the degradation of Rhodamine B (RhB) dye. Three machine learning algorithms (Random Forest, Artificial Neural Network and Xtreme Gradient Boosting) were used in the regression model development. The dataset used in the machine learning and data correlation was generated by Central Composite Rotational Design (CCRD 2²). Regarding the machine learning study, the ANN with structure 3:6:1 showed the best performance as a predictive model (R² = 0.98 and 0.9 for training and testing, RMSE < 5.0), resulting in the 50.37 ± 1.01 % RhB removal at pH 5.7, [RhB] = 200 mg L−1 and [nANA] = 2.75 g L−1 after 180 min under visible light. Feature importance revealed that all parameters (pH, [RhB], [nANA]) were relevant to the response. Therefore, this work confirms the potentiality of machine learning algorithms to develop predictive models as well as a good starting point for the scale-up of advanced oxidation processes.

On the use of hydrodynamic modelling and random forest classifiers for the prediction of hypoxia in coastal lagoons

Hypoxia is one of the fundamental threats to water quality globally, particularly for partially enclosed basins with limited water renewal, such as coastal lagoons. This work proposes the combined use of a machine learning technique, field observations, and data derived from a hydrodynamic and heat exchange numerical model to predict, and forecast up to 10 days in advance, the occurrence of hypoxia in a eutrophic coastal lagoon. The random forest machine learning algorithm is used, training and validating a set of models to classify dissolved oxygen levels in the lagoon. The Orbetello lagoon, in the central Mediterranean Sea (Italy), has provided a test case for assessing the reliability of the proposed methodology. Results proved that the methodology is effective in providing a reliable short-term evaluation of DO levels, with a high resolution in both time and space throughout an entire lagoon. An overall classification accuracy of up to 91 % was found in the models, with a score for identifying the occurrence of severe hypoxia - i.e. hourly DO levels lower than 2 mg/l - of 86 %. The use of predictors extracted from a numerical hydrodynamic model allows us to overcome the intrinsic limitation of machine learning modelling approaches which rely on input data from relatively few, local field measurements, i.e. the inability to capture the spatial heterogeneity of DO distributions, unless several measuring points are available. The methodological approach is proposed for application to similar eutrophic environments.

Prediction of Hearing Help Seeking to Design a Recommendation Module of an mHealth Hearing App: Intensive Longitudinal Study of Feature Importance Assessment.

Mobile health (mHealth) solutions can improve the quality, accessibility, and equity of health services, fostering early rehabilitation. For individuals with hearing loss, mHealth apps might be designed to support the decision-making processes in auditory diagnostics and provide treatment recommendations to the user (eg, hearing aid need). For some individuals, such an mHealth app might be the first contact with a hearing diagnostic service and should motivate users with hearing loss to seek professional help in a targeted manner. However, personalizing treatment recommendations is only possible by knowing the individual's profile regarding the outcome of interest. This study aims to characterize individuals who are more or less prone to seeking professional help after the repeated use of an app-based hearing test. The goal was to derive relevant hearing-related traits and personality characteristics for personalized treatment recommendations for users of mHealth hearing solutions. In total, 185 (n=106, 57.3% female) nonaided older individuals (mean age 63.8, SD 6.6 y) with subjective hearing loss participated in a mobile study. We collected cross-sectional and longitudinal data on a comprehensive set of 83 hearing-related and psychological measures among those previously found to predict hearing help seeking. Readiness to seek help was assessed as the outcome variable at study end and after 2 months. Participants were classified into help seekers and nonseekers using several supervised machine learning algorithms (random forest, naïve Bayes, and support vector machine). The most relevant features for prediction were identified using feature importance analysis. The algorithms correctly predicted action to seek help at study end in 65.9% (122/185) to 70.3% (130/185) of cases, reaching 74.8% (98/131) classification accuracy at follow-up. Among the most important features for classification beyond hearing performance were the perceived consequences of hearing loss in daily life, attitude toward hearing aids, motivation to seek help, physical health, sensory sensitivity personality trait, neuroticism, and income. This study contributes to the identification of individual characteristics that predict help seeking in older individuals with self-reported hearing loss. Suggestions are made for their implementation in an individual-profiling algorithm and for deriving targeted recommendations in mHealth hearing apps.

Ex-ante expected changes in ESG and future stock returns based on machine learning

This study has two primary objectives. Firstly, it enhances the reliability and transparency of machine-learning-based models for predicting future changes in environmental, social and governance (ESG) performance. Secondly, it explores the relationship between ex-ante expected changes in ESG and future stock returns. This study collects 3258 STOXX Europe 600 firm-year observations. In the ESG prediction phase, two machine learning algorithms (logistic regression and random forest) are utilised to develop ESG forecasting models. Hyperparameter optimisation and walk-forward validation techniques are employed to address issues of underestimation and information leakage. The machine-learning-based ESG forecasting models are evaluated using three metrics: accuracy, area under the curve (AUC) and area under the precision-recall curve (AUPR). Subsequently, this study investigates the relationship between ex-ante expected ESG changes and future stock returns using the predicted ESG changes. A positive correlation is found between ex-ante expected ESG changes and future stock returns. The supplementary tests also reveal that this positive relationship is highly and statistically significant among large firms and after the COVID-19 pandemic. Moreover, this study introduces a robust and transparent approach for constructing effective machine-learning-based ESG forecasting models using hyperparameter optimisation and walk-forward validation. Additionally, traditional regression analyses are modernised by incorporating machine-learning-predicted independent variables. Furthermore, the findings provide empirical support for stakeholder, agency and resource-based theories. Finally, practical insights are provided to facilitate ESG-focused investment portfolio decision making.

Classification of deep and shallow groundwater wells based on machine learning in the Hebei Plain North China

Accurately determining the extraction volumes from various aquifers is crucial for effectively managing groundwater overexploitation. A key initial step in quantifying extracted groundwater volumes involves the classification of groundwater wells as either deep or shallow. This study evaluated 881,872 groundwater wells in the Hebei Plain, applying machine learning techniques to classify wells with unknown depths. Through the hydrogeological borehole data, the groundwater wells with known depth are divided into deep wells and shallow wells. Four machine learning algorithms—Random Forest, Support Vector Machine, Logistic Regression, and Naive Bayes—were employed to classify groundwater wells with unknown depths. The accuracy of these models was validated using known-depth well classifications. The results reveal that the Random Forest algorithm exhibited the highest performance among the models, achieving an overall accuracy of 91.23%. According to the Random Forest model, 43.51% of groundwater wells with unknown depths were classified as deep, while 56.49% were classified as shallow. The study also found that wells in areas where salinity exceeds 2 g/L are primarily deep groundwater wells. These findings provide valuable technical insight for groundwater well decommissioning and facilitate the assessment of extracted volumes of deep and shallow groundwater.