Gradient Boosting Regression Research Articles

Spatial Modeling and Mapping of Riverbank Erosion Through the Integration of Machine Learning Application and In-situ Data

Assessing riverbank erosion is crucial for identifying hazard zones and implementing protective measures from potential disasters. Traditional in-situ measurements, though effective, are often costly and time-consuming for large-scale evaluations. This study proposes a methodology integrating machine learning (ML) applications to assess riverbank erosion across an entire area using existing submerged Jet erosion test (JET) measurements. In-situ JETs were used to measure the bank erosion rate at each site, identify influencing factors, and randomly split datasets for training and testing. Four ML techniques, random forest(RF), decision trees(DT), multiple linear regression(MLR), and gradient boosting regressor(GBR), are applied to establish the correlation between riverbank erosion and its influencing factors. The RF model demonstrated the highest accuracy (R2 = 0.94, RMSE = 3.04, and NSE = 0.93) among all algorithms, and the optimal model was applied to predict and map annual erosion rates, which were validated with additional submerged JET data. The proposed methodology can effectively model and map riverbank erosion in similar settings.

Enhanced Predictive Ensemble Learning Models for Compressive Strength of Concrete Modified with Bentonite Clay

This study addresses the logistical challenges of testing bentonite plastic concrete in the field by developing predictive models for compressive strength (CS) using Python-based machine learning techniques. The models, including random forest, adaptive boosting, extreme gradient boosting, and gradient boosting regression tree, were enhanced with forensic-based investigation optimization. A dataset of 285 CS records was used to train and validate these models. SHAP analysis highlighted the impact of various inputs like gravel, bentonite, curing time, and cement on CS. Results show that the GBRT-FBIO model outperformed the others in CS prediction, indicating its superior effectiveness for material science applications.

Combining Sentinel-2 and diverse environmental data largely improved aboveground biomass estimation in China's boreal forests.

Accurately mapping aboveground biomass (AGB) in China's boreal forests is crucial for assessing global carbon stock and formulating forest management strategies but remains challenging as the environmental heterogeneity complicates AGB estimation. Here, we investigated the relative gains of integrating Sentinel-2 and environmental data, as well as synthetic aperture radar (SAR) images to map AGB in China's boreal forests. We used two machine learning algorithms, random forest and gradient boosting regression (GBR), and four dataset combinations to develop the AGB models, then evaluated the AGB map by carrying on uncertainty analysis and comparing it with existing AGB products. Results showed that the GBR model based on Sentinel-2 and environmental data presented the best AGB estimation capability (R2: 0.75, RMSE: 23.60 Mg/ha), while further adding SAR images had negative effects on the model improvement. The Tasseled Cap Distance, short-wave infrared from Sentinel-2, Black dragon fire disturbance, Elevation, and Geographic locations were found to be significant contributors to AGB prediction. Our AGB estimates exhibited moderate to low uncertainty and outperformed other existing AGB maps in China's boreal forests based on independent validation assessment. The AGB distribution presented a noticeable south-north gradient difference, ranging from 3.23 to 346.37 Mg/ha. This study provides new insight into AGB estimation through the integration of Sentinel-2 imagery and multiple environmental data and offers a basis for sustainable management in China's boreal forests.

Comparison of Machine Learning Algorithms for Prediction of Textile Effluent Treatment Efficiency Using Anaerobic Process

ABSTRACTThe prediction of pollutants removal efficiency from the generated effluent of a treatment plant is valuable and can reduce the time, sampling and energy required during performance assessment. The present study aims to predict the effect of different input parameters on the treatment efficiency of the developed microbial‐based anaerobic process for textile effluent using machine leaning algorithms. The decolourisation and chemical oxygen demand (COD) reduction of the treated effluent were predicted on the basis of the three different input parameters pH, COD and colour value of the textile wastewater. The effectiveness of different machine learning algorithms, support vector machines (SVM), random forest (RF), gradient boost regressor (GBR), AdaBoost, extreme gradient boosting (XGB) regressor and voting regressor, were evaluated based on the correlation coefficient (R2) value. The results revealed that the RF achieved the highest accuracy for decolourisation (training data R2: ∼0.85 and test data R2: ∼0.84) as well as COD reduction (training data R2: ∼0.87 and test data R2: ∼0.94) compared to the other algorithms. These results were validated experimentally, confirming that RF can be used as a tool to predict the performance efficiency of a microbial‐based treatment system.

Machine learning assisted screening of nitrogen-doped graphene-based dual-atom hydrogen evolution electrocatalysts

Nitrogen doped graphene-based dual-atom catalysts (NG-DACs) usually exhibit high reactivity for hydrogen evolution reaction (HER). The experimental design of efficient DACs-based HER catalysts, however, is time-consuming and expensive. In this work, a density functional theory combined with machine learning (DFT-ML) strategy is adopted towards the rapid screening of NG-DACs HER catalyst. The adsorption energies of HER intermediates on NG-DACs obtained by DFT calculations and their elemental features were used for the ML database construction. The prediction performance of three ML algorithms was evaluated, and gradient boosted regression (GBR) is found exhibit the best prediction accuracy. Using the HER energetics on Pt (111) as reference, quick screening of the NG-DACs with high HER activity was achieved, resulting to twenty-four potential catalysts. Further selection and validation of the ML-screened catalysts were performed by the hydrogen adsorption Gibbs free energy analysis. Accordingly, four promising NG-DACs HER catalysts were predicted using the proposed DFT-ML based catalyst prediction framework. The DFT-ML strategy offers an promising approach for the screening of new HER electrocatalysts.

Compressive behavior of elliptical concrete-filled steel tubular short columns using numerical investigation and machine learning techniques

This paper presents a non-linear finite element model (FEM) to predict the load-carrying capacity of three different configurations of elliptical concrete-filled steel tubular (CFST) short columns: double steel tubes with sandwich concrete (CFDST), double steel tubes with sandwich concrete and concrete inside the inner steel tube, and a single outer steel tube with sandwich concrete. Then, a parametric and analytical study was performed to explore the influence of geometric and material parameters on the load-carrying capacity of elliptical CFST short columns. Furthermore, the current study investigates the effectiveness of machine learning (ML) techniques in predicting the load-carrying capacity of elliptical CFST short columns. These techniques include Support Vector Regressor (SVR), Random Forest Regressor (RFR), Gradient Boosting Regressor (GBR), XGBoost Regressor (XGBR), MLP Regressor (MLPR), K-nearest Neighbours Regressor (KNNR), and Naive Bayes Regressor (NBR). ML models accuracy is assessed by comparing their predictions with FE results. Among the models, GBR and XGBR exhibited outstanding results with high test R2 scores of 0.9888 and 0.9885, respectively. The study provided insights into the contributions of individual features to predictions using the SHapley Additive exPlanations (SHAP) approach. The results from SHAP indicate that the eccentric loading ratio (e/2a) has the most significant effect on the load-carrying capacity of elliptical CFST short columns, followed by the yield strength of the outer steel tube (\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\:{f}_{yo}$$\\end{document}) and the inner width of the inner steel tube (\\documentclass[12pt]{minimal} \\usepackage{amsmath} \\usepackage{wasysym} \\usepackage{amsfonts} \\usepackage{amssymb} \\usepackage{amsbsy} \\usepackage{mathrsfs} \\usepackage{upgreek} \\setlength{\\oddsidemargin}{-69pt} \\begin{document}$$\\:2{a}_{ii}$$\\end{document}). Additionally, a user interface platform has been developed to streamline the practical application of the proposed ML.

Machine learning-enhanced band gaps prediction for low-symmetry double and layered perovskites.

Density functional theory (DFT) calculations are widely used for material property prediction, but their computational cost can hinder the discovery of novel perovskites. This work explores machine learning (ML) as a faster alternative for predicting band gaps in complex perovskites, focusing on low-symmetry double and layered structures. We employ Support Vector Regression (SVR), Random Forest Regression (RFR), Gradient Boosting Regression (GBR), and Extreme Gradient Boosting (XGBoost) to predict both direct and indirect band gaps. Model performance is evaluated using Mean Absolute Error (MAE), Mean Squared Error (MSE), and R-squared (R²) metrics. Our results reveal SVR as the most effective general model for predicting band gaps in both double and layered perovskites. Interestingly, for double perovskites specifically, XGBoost achieves even higher accuracy when incorporating derivative discontinuity as a feature. Feature importance analysis identifies the standard deviation of valence charges ("Valence (std)") as the most critical factor for band gap prediction across all studied perovskites. This research demonstrates the potential of ML for efficient and accurate band gap prediction in complex perovskites, accelerating material discovery efforts.

Predicting the Multiphotonic Absorption in Graphene by Machine Learning

This study analyzes the nonlinear optical properties exhibited by graphene, focusing on the nonlinear absorption coefficient and the nonlinear refractive index. The evaluation was conducted using the Z-scan technique with a 532 nm wavelength laser at various intensities. The nonlinear optical absorption and the nonlinear optical refractive index were measured. Four machine learning models, including linear regression, decision trees, random forests, and gradient boosting regression, were trained to analyze how the nonlinear optical absorption coefficient varies with variables such as spot radius, maximum energy, and normalized minimum transmission. The models were trained with synthetic data and subsequently validated with experimental data. Decision tree-based models, such as random forests and gradient boosting regression, demonstrated superior performance compared to linear regression, especially in terms of mean squared error. This work provides a detailed assessment of the nonlinear optical properties of graphene and highlights the effectiveness of machine learning methods in this context.

Optimizing pyrolysis and Co-Pyrolysis of plastic and biomass using Artificial Intelligence

The rapid increase in biomass and plastic waste poses significant environmental challenges. Co-pyrolysis of biomass with plastic wastes offers a promising avenue for sustainable waste management and renewable energy generation. This study covers several novel aspects: First, it investigates the impacts of feedstock composition and operating conditions in pyrolysis (individual feedstock) and co-pyrolysis (biomass and plastic wastes). The study reveals that synergistic effects, specifically improved yields and optimized temperature, exist in the co-pyrolysis of biomass and plastic wastes compared to individual feedstock. Secondly, a suitable blended machine learning predictive model (with Random Forest, Gradient Boosting Regressor, and XGBoost) and robust optimization framework are developed to address model accuracy, non-linear interactions, and uncertainties in pyrolysis such as temperature, heating rate, and biomass-to-plastic ratio. This study predicts the bio-oil yield quantitatively (amount) and qualitatively (composition) with high accuracy (R2 > 0.97). Thirdly, key factors contributing to yield include plastic content (18 %) and biomass type (13 %) have been identified through Gini feature importance and Shapley Additive Explanation (SHAP) analysis. Furthermore, multi-objective optimization techniques reveal the most optimal bio-oil yield under specific conditions, supported by uncertainty analysis, which confines bio-oil yield to a range of 30–50 %. Finally, it also demonstrates a case study to find the optimal bio-oil yield and quality conditions using co-pyrolysis of local resources, i.e., biomass (wood and bagasse) and plastic wastes. The case study suggests optimal conditions like > 50 °C heating rate, <50 min pyrolysis time, and > 60 % plastic content in a blend of wood and HDPE. This study assists industries and policymakers to assess and understand the viability of co-pyrolysis, optimal design parameters, and process impacts.

Integration of ML methods with CR model-based optical diagnostic for the estimation of electron temperature in Ga laser produced plasma

Accelerated diagnostic of plasma plays a significant role in controlling and optimizing plasma-mediated processing, particularly for plasma with higher temporal and spatial gradients, such as laser produced plasma (LPP). In the present work, two advanced machine learning (ML) algorithms, random forest regression, and gradient boosting regression are integrated with noninvasive collisional radiative (CR) model-based optical diagnostics to facilitate accurate diagnostics. A comprehensive fine-structure resolved CR model framework is developed by incorporating our consistent cross section data obtained from the Relativistic Distorted Wave method [Suresh et al., “Fully relativistic distorted wave calculations of electron impact excitation of gallium atom: Cross sections relevant for plasma kinetic modelling,” Spectrochim. Acta B: At. Spectrosc. 213, 106860 (2024)]. An extensive dataset of CR model simulated intensities is created to train and test the ML methods. The present CR model is applied to characterize the Gallium LPP coupling with the optical emission spectroscopic measurements of Guo et al. [“Time-resolved spectroscopy analysis of Ga atom in laser induced plasma,” Laser Phys. 19, 1832–1837 (2009)] at different delay times. Further, a detailed correlation study of the line intensity ratios is performed to observe the qualitative behavior of the plasma parameters. The electron temperature results obtained from the CR model, ML, and line ratio methods were compared and found to be in excellent agreement. Overall, the present study demonstrates diagnostic approaches that can benefit the LPP community significantly by providing a rapid understanding of the plasma behavior across various operating conditions.

Daily Runoff Forecasting Using Novel Optimized Machine Learning Methods

Accurate runoff forecasting is crucial for effective water resource management, yet existing models often face challenges due to the complexity of hydrological systems. This study addresses these challenges by introducing a novel bio-inspired metaheuristic algorithm, Artificial Rabbits Optimization (ARO), integrated with various machine learning (ML) models for runoff forecasting in the Carson and Chehalis River basins. The ARO-ML models are rigorously evaluated, demonstrating their superior ability to enhance forecasting accuracy compared to widely used ML techniques, including gradient boosting (GB), multi-layer perceptron (MLP), and K-nearest neighbor regression (KNN). In the Carson River, the GB model achieves the highest forecasting accuracy, which is significantly improved by ARO, resulting in a 24.8% reduction in root mean square error (RMSE). The MLP model also benefits notably from ARO, with RMSE improvements of 4.8% and a substantial 48.9% reduction in mean absolute error (MAE). The ARO-optimized KNN model shows exceptional performance, surpassing the baseline KNN during testing. In the Chehalis River Basin, ARO integration leads to substantial improvements in both GB and MLP models across various performance metrics during training and testing. Crucially, the results indicate that in both river basins, rainfall in preceding days has the most significant influence on streamflow forecasts, followed by air temperature. This study advances hydrological forecasting by validating the efficacy of ARO-ML models, offering a robust framework for enhanced predictive accuracy and providing deeper insights into the key drivers of streamflow in diverse hydrological contexts.

Effective Machine Learning Techniques for Dealing with Poor Credit Data

Credit risk is a crucial component of daily financial services operations; it measures the likelihood that a borrower will default on a loan, incurring an economic loss. By analysing historical data for assessment of the creditworthiness of a borrower, lenders can reduce credit risk. Data are vital at the core of the credit decision-making processes. Decision-making depends heavily on accurate, complete data, and failure to harness high-quality data would impact credit lenders when assessing the loan applicants’ risk profiles. In this paper, an empirical comparison of the robustness of seven machine learning algorithms to credit risk, namely support vector machines (SVMs), naïve base, decision trees (DT), random forest (RF), gradient boosting (GB), K-nearest neighbour (K-NN), and logistic regression (LR), is carried out using the Lending Club credit data from Kaggle. This task uses seven performance measures, including the F1 Score (recall, accuracy, and precision), ROC-AUC, and HL and MCC metrics. Then, the harnessing of generative adversarial networks (GANs) simulation to enhance the robustness of the single machine learning classifiers for predicting credit risk is proposed. The results show that when GANs imputation is incorporated, the decision tree is the best-performing classifier with an accuracy rate of 93.01%, followed by random forest (92.92%), gradient boosting (92.33%), support vector machine (90.83%), logistic regression (90.76%), and naïve Bayes (89.29%), respectively. The classifier is the worst-performing method with a k-NN (88.68%) accuracy rate. Subsequently, when GANs are optimised, the accuracy rate of the naïve Bayes classifier improves significantly to (90%) accuracy rate. Additionally, the average error rate for these classifiers is over 9%, which implies that the estimates are not far from the actual values. In summary, most individual classifiers are more robust to missing data when GANs are used as an imputation technique. The differences in performance of all seven machine learning algorithms are significant at the 95% level.

A novel bi-level optimal strategy-assisted design boosting discovery of specific surface area of perovskite oxide for effective photocatalysis

Due to the inherent property of materials, photocatalysis plays a crucial role in diverse applications, such as the degradation of organic pollutants, fuel production, and nitrogen fixation. Perovskites, in particular, are remarkable for their high light absorption rate and energy conversion efficiency within the visible light range, thereby positioning them as promising photocatalysts. Given that specific surface area (SSA) is a critical metric for evaluating photocatalytic performance, accurately and quickly predicting the SSA of perovskites has emerged as a key area of technical research. Therefore, it is essential to explore the relationship between potential materials descriptors and SSA, which can significantly expedite the discovery of photocatalyzed perovskite materials with high SSA. In this study, we proposed a bi-level optimization method that combines descriptors identified strategy and model optimization strategy for predicting the SSA of unknown perovskite candidates, thereby elucidating the relationship between material features and SSA. Initially, at the upper level, the Gradient Boosting Regression combined with Recursive Feature Elimination (RFE-GBR) method was employed to identify crucial features that are closely correlated with the SSA of photocatalyzed perovskites. Subsequently, the Sure Independence Screening.Sparsifying Operator (SISSO) was utilized to develop new descriptors that exhibited strong correlations with SSA. Based on refined descriptors identifiied at the upper level, at the lower level, the optimal model strategy was obtained by Improved Osprey Optimization Algorithm (IOOA) using golden sine strategy and Gradient Boosting Regression (GBR) regression to tune the key parameters of GBR for optimizing SSA. The optimal designing method achieved remarkable predictive performance with R2 (R-squared) score of 0.90 and MAE (Mean Absolute Error) of 3.97 m2/g, along with 5-fold cross-validation. Then this method was applied to predict the SSA of 300 unidentified ABO3 perovskites. Comparison with experimental data revealed that our approach significantly accelerated the development of perovskite materials with superior photocatalytic properties, showing the potential of our model in advancing the field of photocatalysis.

Clinical and socioeconomic predictors of hospital use and emergency department visits among children with medical complexity: A machine learning approach using administrative data

Objectives The primary objective of this study was to identify clinical and socioeconomic predictors of hospital and ED use among children with medical complexity within 1 and 5 years of an initial discharge between 2010 and 2013. A secondary objective was to estimate marginal associations between important predictors and resource use. Methods This retrospective, population-cohort study of children with medical complexity in Alberta linked administrative health data with Canadian census data and used tree-based, gradient-boosted regression models to identify clinical and socioeconomic predictors of resource use. Separate analyses of cumulative numbers of hospital days and ED visits modeled the probability of any resource use and, when present, the amount of resource use. We used relative importance in each analysis to identify important predictors. Results The analytic sample included 11 105 children with medical complexity. The best short- and long-term predictors of having a hospital stay and number of hospital days were initial length of stay and clinical classification. Initial length of stay, residence rurality, and other socioeconomic factors were top predictors of short-term ED use. The top predictors of ED use in the long term were almost exclusively socioeconomic, with rurality a top predictor of number of ED visits. Estimates of marginal associations between initial length of stay and resource use showed that average number of hospital days increases as initial length of stay increases up to approximately 90 days. Children with medical complexity living in rural areas had more ED visits on average than those living in urban or metropolitan areas. Conclusions Clinical factors are generally better predictors of hospital use whereas socioeconomic factors are more predictive of ED use among children with medical complexity in Alberta. The results confirm existing literature on the importance of socioeconomic factors with respect to health care use by children with medical complexity.

UAV-Based Multispectral Winter Wheat Growth Monitoring with Adaptive Weight Allocation

Comprehensive growth index (CGI) more accurately reflects crop growth conditions than single indicators, which is crucial for precision irrigation, fertilization, and yield prediction. However, many current studies overlook the relationships between different growth parameters and their varying contributions to yield, leading to overlapping information and lower accuracy in monitoring crop growth. Therefore, this study focuses on winter wheat and constructs a comprehensive growth monitoring index (CGIac), based on adaptive weight allocation of growth parameters’ contribution to yield, using data such as leaf area index (LAI), soil plant analysis development (SPAD) values, plant height (PH), biomass (BM), and plant water content (PWC). Using UAV data on vegetation indices, feature selection was performed using the Elastic Net. The growth inversion model was then constructed using machine learning methods, including linear regression (LR), random forest (RF), gradient boosting (GB), and support vector regression (SVR). Based on the optimal growth inversion model for winter wheat, spatial distribution of wheat growth in the study area is obtained. The findings demonstrated that CGIac outperforms CGIav (constructed using equal weighting) and CGIcv (built using the coefficient of variation) in yield correlation and prediction accuracy. Specifically, the yield correlation of CGIac improved by up to 0.76 compared to individual indices, while yield prediction accuracy increased by up to 23.14%. Among the evaluated models, the RF model achieved the best performance, with a coefficient of determination (R2) of 0.895 and a root mean square error (RMSE) of 0.0058. A comparison with wheat orthophotos from the same period confirmed that the inversion results were highly consistent with actual growth conditions in the study area. The proposed method significantly improved the accuracy and applicability of winter wheat growth monitoring, overcoming the limitations of single parameters in growth prediction. Additionally, it provided new technological support and innovative solutions for regional crop monitoring and precision farming operations.

Evaluating the performances of SVR and XGBoost for short-range forecasting of heatwaves across different temperature zones of India

This research aims to forecast maximum temperatures and the frequency of heatwave days across four different temperature zones (Zone 1, 2, 3 and 4) in India. These four zones are categorized based on the 30-year average maximum temperatures (T30AMT) during the summer months of April, May, and June (AMJ). Two Machine Learning (ML) algorithms eXtreme Gradient Boosting (XGBoost) and Support Vector Regression (SVR) are employed to achieve this goal. The study utilizes nine key atmospheric variables namely air temperature, geopotential height, relative humidity, U-wind, V-wind, soil moisture, solar radiation, sea surface temperature, and mean sea level pressure at a daily scale spanning from 1991 to 2020 for the months of March, April, May, and June as predictors. The India Meteorological Department daily maximum temperature data spanning from 1991 to 2020 for the months of AMJ serves as the predictands. ML models are developed using spatially averaged atmospheric variables and daily maximum temperature across the grids falling within each temperature zone. Results indicate that for a 7-day lead time, SVR outperforms XGBoost in Zone-1 (T30AMT > 38 °C) and Zone-2 (T30AMT: 35.01 °C–38 °C) by more accurately capturing peak temperatures during training and testing. Conversely, for a 15-day lead time in Zone-1, XGBoost better predicts temperature peaks in both phases. In Zone-3 (T30AMT: 30 °C–35 °C) and Zone-4 (T30AMT < 30 °C) for both lead times, the performance of both models decline, indicating models and input variables are more effective in predicting higher temperatures typical of Zone-1 and 2 but less so in Zone-3 and 4. In a nutshell, the study attempts to highlight the capability of advanced ML techniques combined with spatial climate data to enhance the prediction of extreme heatwave events. These insights can aid in heatwave preparedness, climate management, and adaptation strategies for different Indian temperature zones.

Prediction of gastrointestinal bleeding hospitalization risk in hemodialysis using machine learning

BackgroundGastrointestinal bleeding (GIB) is a clinical challenge in kidney failure. INSPIRE group assessed if machine learning could determine a hemodialysis (HD) patient’s 180-day GIB hospitalization risk.MethodsAn eXtreme Gradient Boosting (XGBoost) and logistic regression model were developed using an HD dataset in United States (2017–2020). Patient data was randomly split (50% training, 30% validation, and 20% testing). HD treatments ≤ 180 days before GIB hospitalization were classified as positive observations; others were negative. Models considered 1,303 exposures/covariates. Performance was measured using unseen testing data.ResultsIncidence of 180-day GIB hospitalization was 1.18% in HD population (n = 451,579), and 1.12% in testing dataset (n = 38,853). XGBoost showed area under the receiver operating curve (AUROC) = 0.74 (95% confidence interval (CI) 0.72, 0.76) versus logistic regression showed AUROC = 0.68 (95% CI 0.66, 0.71). Sensitivity and specificity were 65.3% (60.9, 69.7) and 68.0% (67.6, 68.5) for XGBoost versus 68.9% (64.7, 73.0) and 57.0% (56.5, 57.5) for logistic regression, respectively. Associations in exposures were consistent for many factors. Both models showed GIB hospitalization risk was associated with older age, disturbances in anemia/iron indices, recent all-cause hospitalizations, and bone mineral metabolism markers. XGBoost showed high importance on outcome prediction for serum 25 hydroxy (25OH) vitamin D levels, while logistic regression showed high importance for parathyroid hormone (PTH) levels.ConclusionsMachine learning can be considered for early detection of GIB event risk in HD. XGBoost outperforms logistic regression, yet both appear suitable. External and prospective validation of these models is needed. Association between bone mineral metabolism markers and GIB events was unexpected and warrants investigation.Trial registrationThis retrospective analysis of real-world data was not a prospective clinical trial and registration is not applicable.Graphical

Evaluation of machine learning models for the accelerated prediction of density functional theory calculated 19F chemical shifts based on local atomic environments

The introduction of fluorine in compounds plays a crucial role in drug development as it greatly influences their final pharmacokinetic and dynamic properties. Due to the prevalence of fluorine in FDA-approved drugs in recent years, identifying the mechanisms driving their chemical transformations has become crucial in the drug discovery landscape. 19F NMR spectroscopy is a powerful analytical technique that allows for the examination of fluorine-containing compounds, offering valuable information about their structure, dynamics, and reactivity. NMR spectra can be interpreted through the leveraging of Density Functional Theory (DFT). However, the screening of compounds and discovery of feasible drug candidates is limited due to its computational cost. Here, we present a machine learning approach to accelerate the prediction of DFT-calculated 19F NMR chemical shifts. The fluorine atoms’ features in the models were derived from their local three-dimensional environments, representing their neighboring atoms within a radius of n Å away from the given fluorine atom in the compound. A comparative analysis of thirteen regression models was conducted using features extracted from 501 fluorinated compounds in our laboratory’s chemical inventory. Among the models, Gradient Boosting Regression (GBR) exhibited the highest performance, achieving a mean absolute error of 3.31 ppm with a local environment radius of 3 Å. This demonstrates a comparable accuracy to DFT calculations while reducing computational time from several hundred seconds to milliseconds. 3 Å was also found to be the most optimal radius across all models when encoding features for local atomic environments.

Optimizing H2 production from biomass: A machine learning-enhanced model of supercritical water gasification dynamics

Supercritical water gasification (SCWG) is recognized as an efficient technology for biomass conversion, demonstrating substantial potential in the sustainable energy sector. This paper focuses on the H2 production by SCWG of biomass. The objective is to develop an accurate gasification kinetic model that can facilitate the optimization of industrial reactor designs. A series of SCWG experiments is conducted in a batch reactor system under temperature of 500–600 °C and residence time of 1–20 min. Based on the experimental results, a hybrid-driven model for SCWG reaction is established. Firstly, experimental data is utilized to delineate SCWG reaction mechanistic model and forecast gas yields with an acceptable error margin. Subsequently, a Wasserstein Generative Adversarial Network Gradient Boosting Regression Grid Search (WGAN-GBR-GRID) model is applied to acquire knowledge of the predictive errors from the mechanistic model and to develop a hybrid model. The hybrid-driven model significantly enhances the average relative prediction error from 15.56 % to 0.02 % when compared to the mechanistic model. This result underscores the potential of the hybrid model in optimizing SCWG technology and the Shapley Additive exPlanations (SHAP) values in feature analysis shows the possible shortcomings of mechanistic model, thereby providing a new perspective for SCWG reaction dynamics modeling.

Machine Learning-Based Prediction of Aircraft Taxi-In Time: A Case Study at Beijing Capital International Airport

Accurate prediction of flight taxi-in time has a significant meaning in allocating aircraft support resources reasonably and improving airport surface movement efficiency. It can effectively overcome the deficiency of extensive aircraft arrival time prediction in major airports currently. Taking Beijing Capital International Airport as the research object, we analyzed the influence factors of taxi-in time and constructed the feature set; Then we applied linear regression, K-nearest neighbor, support vector regression, decision tree, random forest and gradient boosting regression tree, which were widely used in the prediction of taxi-out time, to predict the taxi-in time. The results show that the prediction accuracy of the six machine learning models is over 90%within \pm3minutes, it means that the construction of feature set and the selection of models are effective; The gradient boosting regression tree model has the best performance based on the prediction results and model fitting evaluation results; The surface traffic flow features have the largest contribution to the prediction model, and the newly introduced cross-regional feature has more contribution to the prediction model than most traditional features, according to the prediction results of gradient boosting regression tree.