Gradient Boosting Regression Tree Model Research Articles

Hybrid machine learning model based predictions for properties of poly(2-hydroxyethyl methacrylate)-poly(vinyl alcohol) composite cryogels embedded with bacterial cellulose

Supermacroporous composite cryogels with enhanced adjustable functionality have received extensive interest in bioseparation, tissue engineering, and drug delivery. However, the variations in their components significantly impactfinal properties. This study presents a two-step hybrid machine learning approach for predicting the properties of innovative poly(2-hydroxyethyl methacrylate)-poly(vinyl alcohol) composite cryogels embedded with bacterial cellulose (pHEMA-PVA-BC) based on their compositions. By considering the ratios of HEMA (1.0–22.0 wt%), PVA (0.2–4.0 wt%), poly(ethylene glycol) diacrylate (1.0–4.5 wt%), BC (0.1–1.5 wt%), and water (68.0–96.0 wt%) as investigational variables, overlay sampling uniform design (OSUD) was employed to construct a high-quality dataset for model development. The random forest (RF) model was used to classify the preparation conditions. Then four models of artificial neural network, RF, gradient boosted regression trees (GBRT), and XGBoost were developed to predict the basic properties of the composite cryogels. The results showed that the RF model achieved an accurate three-class classification of preparation conditions. Among the four models, the GBRT model exhibited the best predictive performance of the basic properties, with the mean absolute percentage error of 16.04 %, 0.85 %, and 2.44 % for permeability, effective porosity, and height of theoretical plate (1.0 cm/min), respectively. Characterization results of the representative pHEMA-PVA-BC composite cryogel showed an effective porosity of 81.01 %, a permeability of 1.20 × 10−12 m2, and a range of height of theoretical plate between 0.40–0.49 cm at flow velocities of 0.5–3.0 cm/min. These indicate that the pHEMA-PVA-BC cryogel was an excellent material with supermacropores, low flow resistance and high mass transfer efficiency. Furthermore, the model output demonstrates that the alteration of the proportions of PVA (0.2–3.5 wt%) and BC (0.1–1.5 wt%) components in composite cryogels resulted in significant changes in the material basic properties. This work represents an attempt to efficiently design and prepare target composite cryogels using machine learning and providing valuable insights for the efficient development of polymers.

Predicting the properties of perovskite materials by improved compositionally restricted attention-based networks and explainable machine learning

Understanding the unique properties of perovskite materials is crucial in advancing solar energy technologies. Factors like heat of formation and bandgap significantly influence the light absorption capability and stability of perovskite solar cells. However, it is time-consuming and labor-intensive to obtain the properties of perovskites using traditional experimental or high-throughput computational methods. As a prospective method, machine learning can find regularities in the given training data and give accurate prediction results. In this article, we use deep learning models based on attention mechanisms and elemental features to predict the heat of formation and bandgap of perovskite materials. Random Forest and Gradient Boosted Regression Tree models have also been used for interpretable predictions of properties. The compositionally restricted attention-based network was improved by introducing a densely connected network and optimizing the network structure to increase data processing capabilities. The experiment results show that the mean absolute errors of the heat of formation and bandgap on the test sets are decreased by 5.77% and 3.37% respectively. The optimized model also shows better performance when used for classification tasks. In addition, we use the gradient boosting regression tree model and the shapley additive explanations tool to conduct an interpretable analysis, explaining the impact of different features on the predictions of the properties.

A tree-based automated machine learning approach of the obstructed view factor of thermal radiation in nuclear pebble beds

For the nuclear pebble bed of the high temperature gas-cooled reactor (HTGR), a tree-based automated machine learning approach is developed to discuss the complicated thermal radiation behaviors. The AutoML model for calculating the obstructed view factor between any two particles in the pebble bed includes the gradient boosting regression tree model with the fine-tuned hyperparameters and the analytical base model. Using the discrete packing of the HTR-PM nuclear reactor and the leaped Halton sequence, a large dataset of the view factor under different conditions is generated to train the AutoML model. On the same platform, numerical results show that AutoML model is approximately 5 × 105 times faster than the traditional approach. The Pareto front of the AutoML model indicates that the mean squared error decreases with the model complexity until it reaches the optimal solution. Then, the trained AutoML model is applied to the engineering pebble bed. It is shown that the average radiation exchange factor near the wall is lower than that in the bulk region. In addition, the packing height is also an important parameter for evaluating the radiative behaviors. In particular, when the height of the nuclear pebble bed is less than 20 times particle diameter, it is necessary to consider the contribution of the packing height to the radiation effective thermal conductivity.

Predicting Band Gap of Transition Metal Trihalides using Machine Learning

Machine Learning (ML) is a tool that finds pattern in data sets for analysis and prediction and is very useful in various sectors such as materials discovery, predicting material properties, medical science, and data science etc. From predicting lattice parameters, stability, band gap, and many other material properties with low cost and good accuracy, ML is becoming popular among material scientists. Recently, a new compound in a two-dimensional (2D) realm (transition metal trihalides) is getting immense interest due to its intriguing properties such as magnetic order in lower dimensions. In this work, we have selected transition metal trihalides to predict band gaps using ML. For the dataset, we have used some compound and elemental properties extracted from the materials project database and periodic table for the features set, and band gaps within the range of 0-3 eV for target variables. We performed the least absolute shrinkage and selection operator (LASSO) for feature selection and selected five features from a set of 24 features. After training and testing our data in four types of ML algorithms (support vector regression (SVR), kernel ridge regression (KRR), gradient boosted regression tree (GBRT), and random forest regression (RFR)), the GBRT model is found to be best with the lowest mean absolute error (MAE) < 0.27 eV and root mean squared error (RMSE) < 0.34 eV.

Machine learning driven prediction of lattice constants in transition metal dichalcogenides

Machine learning represents an emerging branch of artificial intelligence, centering on the enhancement of algorithms in computer programs through the utilization of data and the accumulation of research-driven knowledge. The requirement for artificial intelligence in materials science is essential due to the significant need for innovative high-performance materials on a large scale. In this report, the gradient boosting regression tree model of machine learning was applied to predict the lattice constants of cubic and trigonal MX2 systems (M=transition metal and X=chalcogen atoms). The theoretical/experimental values of the materials were compared to the predicted values to calculate the standard errors such as RMSE (root mean square error) and MAE (mean absolute error). The features used to predict lattice constants were ionic radius, lattice angles, bandgap, formation energy, total magnetic moment, density and oxidation states. The features versus contribution barplot has been drawn to reveal the contribution level of each parameter in the degree of [0,1] to obtain the predictions. This report provides a precise account of the prediction methodology for lattice parameters of the transition metal dichalcogenides family, a process that was previously not reported.

A comprehensive pyrolysis study of sorghum and reed stalk: TG/IR/MS analysis and machine learning-assisted kinetic prediction

This work focused on investigating the pyrolysis characteristics of sorghum stalk (SS) and reed stalk (RS), whose potentiality as value-added feedstocks for thermochemical conversion is largely unexplored. The experimental method, thermogravimetric-infrared-mass spectrometry, was applied to reveal the thermal decomposition process, evaluate the kinetic and thermodynamic parameters, and identify the gaseous products, which could provide fundamental data and valuable information for bioenergy utilization and optimization of reactors. The average activation energy values of SS and RS by isoconversional methods are estimated to be 185.0–210.4 kJ/mol and 192.6–197.0 kJ/mol. Though kinetic parameters, especially activation energy values, play a crucial role in understanding reaction mechanism and guiding further applications, conventional methods are time-consuming and labor-intensive. In the current work, a gradient boosting regression tree model was successfully constructed to achieve fast prediction of activation energy values with a competitive accuracy of R2 > 0.85 without repetitive experiments. In addition, the established model has showed good generalization ability in new datasets, which reveals the great potential of the machine learning approach in the pyrolysis kinetic field.

Mechanical properties prediction of blast furnace slag and fly ash-based alkali-activated concrete by machine learning methods

In this paper, 871 data were collected from literature and trained by the 4 representative machine learning methods, in order to build a robust compressive strength predictive model for slag and fly ash based alkali activated concretes. The optimum models of each machine learning method were verified by 4 validation metrics and further compared with an empirical formula and experimental results. Besides, a literature study was carried out to investigate the connection between compressive strength and other mechanical characteristics. As a result, the gradient boosting regression trees model and several predictive formulas were eventually proposed for the prediction of the mechanical behavior including compressive strength, elastic modulus, splitting tensile strength, flexural strength, and Poisson’s ratio of BFS/FA-AACs. The importance index of each parameter on the strength of BFS/FA-AACs was elaborated as well.

Forecasting daily extreme temperatures in Chinese representative cities using artificial intelligence models

Accurate daily extreme temperature forecasts are important for public safety. However, traditional Numerical Weather Prediction (NWP) methods for temperature forecasting become increasingly costly and time-consuming as forecast resolution improves. To overcome these challenges, recent studies have embraced artificial intelligence (AI) techniques. In this paper, we select five AI models: Multiple linear regression, Support Vector Regressor, Gradient Boosting Regression Tree (GBRT), Long Short-Term Memory, and multilayer perceptron (MLP) to forecast extreme temperatures in nine representative Chinese cities, which are Harbin, Xian, Beijing, and Lhasa in the North and Shanghai, Guiyang, Guangzhou, Shenzhen, and Haikou in the South. To assess the performances, we compare AI models with two benchmarks: the persistence method by using the previous day's temperature extremes directly, and the prediction from the NWP method of the Global Forecast System (GFS) provided by the National Centers for Environmental Prediction. Among the 7 models forecasting daily extreme temperatures in nine cities, the MLP model outperforms the best, followed by the GBRT model. Compared with the persistence forecast, all 5 AI models and GFS show forecasting ability for daily maximum temperature. However, the GFS performs worse than the persistence model for daily minimum temperature. Specifically, for the MLP model, the average Mean Absolute Errors (MAEs) for forecasting daily maximum and minimum temperatures in the 9 cities are 1.61 °C and 1.14 °C, reducing the GFS model's MAEs for daily maximum and minimum temperatures forecasts by 15% and 45%, respectively. Overall, AI models show promise in enhancing temperature forecasting accuracy.

Prediction of Daily Water Consumption in Residential Areas Based on Meteorologic Conditions—Applying Gradient Boosting Regression Tree Algorithm

A more accurate way of water consumption forecasting can be used to help people develop a scheduling plan of water workers more targeting; therefore, this paper aims to establish a forecast model of daily water consumption based on meteorological conditions. At present, most studies of daily water consumption forecasts focus on historical data or single water use influencing factors; moreover, daily water consumption could be influenced by meteorologic conditions. The influence of complex meteorology factors on water consumption is analyzed based on a gradient-boosted regression tree (GBRT) model. The correlation of 10 meteorologic factors has been discussed and divided into 5 categories, including temperature factor, pressure factor, precipitation factor, sunshine factor, and wind factor. Through the GBRT algorithm, the daily water consumption of residential area could be predicted with a maximum error of ±8%. The results show that the average ground temperature (the feature importance accounts for 81% of the total) has the greatest impact on the daily water consumption of the residential community, followed by the somatosensory temperature (the feature importance accounts for 7% of the total). The method can provide the daily water consumption of water consumption nodes with higher precision for municipal water supply network model accuracy. It also provides a reference for water utility operation schemes and urban development planning.

Predicting optimal endotracheal tube size and depth in pediatric patients using demographic data and machine learning techniques.

Use of endotracheal tubes (ETTs) with appropriate size and depth can help minimize intubation-related complications in pediatric patients. Existing age-based formulae for selecting the optimal ETT size present several inaccuracies. We developed a machine learning model that predicts the optimal size and depth of ETTs in pediatric patients using demographic data, enabling clinical applications. Data from 37,057 patients younger than 12 years who underwent general anesthesia with endotracheal intubation were retrospectively analyzed. Gradient boosted regression tree (GBRT) model was developed and compared with traditional age-based formulae. The GBRT model demonstrated the highest macro-averaged F1 scores of 0.502 (95% CI 0.486, 0.568) and 0.669 (95% CI 0.640, 0.694) for predicting the uncuffed and cuffed ETT size (internal diameter [ID]), outperforming the age-based formulae that yielded 0.163 (95% CI 0.140, 0.196, P < 0.001) and 0.392 (95% CI 0.378, 0.406, P < 0.001), respectively. In predicting the ETT depth (distance from tip to lip corner), the GBRT model showed the lowest mean absolute error (MAE) of 0.71 cm (95% CI 0.69, 0.72) and 0.72 cm (95% CI 0.70, 0.74) compared to the age-based formulae that showed an error of 1.18 cm (95% CI 1.16, 1.20, P < 0.001) and 1.34 cm (95% CI 1.31, 1.38, P < 0.001) for uncuffed and cuffed ETT, respectively. The GBRT model using only demographic data accurately predicted the ETT size and depth. If these results are validated, the model may be practical for predicting optimal ETT size and depth for pediatric patients.

Effects of microplastics on denitrification and associated N2O emission in estuarine and coastal sediments: insights from interactions between sulfate reducers and denitrifiers

Global estuarine and coastal zones are facing severe microplastics (MPs) pollution. Sulfate reducers (SRB) and denitrifiers (DNB) are two key functional microorganisms in these zones, exhibiting intricate interactions. However, whether and how MPs modulate the interactions between SRB and DNB, with implications for denitrification and associated N2O emissions, remains poorly understood. Here, we simultaneously investigated the spatial response patterns of SRB-DNB interactions and denitrification and associated N2O emissions to different MPs exposure along an estuarine gradient in the Yangtze Estuary. Spatial responses of denitrification to polyvinyl chloride (PVC) and polyadipate/butylene terephthalate (PBAT) MPs exposure were heterogeneous, while those of N2O emissions were not. Gradient-boosted regression tree and multiple regression model analyses showed that sulfide, followed by nitrate (NO3−), controlled the response patterns of denitrification to MPs exposure. Further mechanistic investigation revealed that exposure to MPs resulted in a competitive and toxic (sulfide accumulation) inhibition of SRB on DNB, ultimately inhibiting denitrification at upstream zones with high sulfide but low NO3− levels. Conversely, MPs exposure induced a competitive inhibition of DNB on SRB, generally promoting denitrification at downstream zones with low sulfide but high NO3− levels. These findings advance the current understanding of the impacts of MPs on nitrogen cycle in estuarine and coastal zones, and provide a novel insight for future studies exploring the response of biogeochemical cycles to MPs in various ecosystems.

Accelerating the Discovery of Hybrid Perovskites with Targeted Band Gaps via Interpretable Machine Learning.

The band gap of hybrid organic-inorganic perovskites (HOIPs) is a key factor affecting the light absorption characteristics and thus the performance of perovskite solar cells (PSCs). However, band gap engineering, using experimental trial and error and high-throughput density functional theory calculations, is blind and costly. Therefore, it is critical to statistically identify the multiple factors influencing band gaps and to rationally design perovskites with targeted band gaps. This study combined feature engineering, the gradient-boosted regression tree (GBRT) algorithm, and the genetic algorithm-based symbolic regression (GASR) algorithm to develop an interpretable machine learning (ML) strategy for predicting the band gap of HOIPs accurately and quantitatively interpreting the factors affecting the band gap. Seven best physical features were selected to construct a GBRT model with a root-mean-square error of less than 0.060 eV, and the most important feature is the electronegativity difference between the B-site and the X-site (χB-X). Further, a mathematical formula (Eg = χB-X2 + 0.881χB-X) was constructed with GASR for a quantitative interpretation of the band gap influence patterns. According to the ML model, the HOIP MA0.23FA0.02Cs0.75Pb0.59Sn0.41Br0.24I2.76 was obtained, with a suitable band gap of 1.39 eV. Our proposed interpretable ML strategy provides an effective approach for developing HOIP structures with targeted band gaps, which can also be applied to other material fields.

Estimation of grassland height using optical and SAR remote sensing data

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.

A New Combined Prediction Model for Ultra-Short-Term Wind Power Based on Variational Mode Decomposition and Gradient Boosting Regression Tree

Wind power is an essential component of renewable energy. It enables the conservation of conventional energy sources such as coal and oil while reducing greenhouse gas emissions. To address the stochastic and intermittent nature of ultra-short-term wind power, a combined prediction model based on variational mode decomposition (VMD) and gradient boosting regression tree (GBRT) is proposed. Firstly, VMD is utilized to decompose the original wind power signal into three meaningful components: the long-term component, the short-term component, and the randomness component. Secondly, based on the characteristics of these three components, a support vector machine (SVM) is selected to predict the long-term and short-term components, while gated recurrent unit-long short-term memory (GRU-LSTM) is employed to predict the randomness component. Particle swarm optimization (PSO) is utilized to optimize the structural parameters of the SVM and GRU-LSTM combination for enhanced prediction accuracy. Additionally, a GBRT model is employed to predict the residuals. Finally, the rolling predicted values of the three components and residuals are aggregated. A deep learning framework using TensorFlow 2.0 has been built on the Python platform, and a dataset measured from a wind farm has been utilized for learning and prediction. The comparative analysis reveals that the proposed model exhibits superior short-term wind power prediction performance, with a mean squared error, mean absolute error, and coefficient of determination of 0.0244, 0.1185, and 0.9821, respectively.

Machine learning models for occurrence form prediction of heavy metals in tailings

ABSTRACT Modern mining and metal ore smelting produce vast tailings, increasing heavy metal pollution. The study of heavy metal occurrence forms is a promising way to remediate contaminated tailings while minimizing environmental damage. However, laboratory measurements of heavy metal occurrence forms are complex and time-consuming, so a fast and accurate identification method is urgently needed. This study used gradient boosting regression tree (GBRT) approaches to predict heavy metal occurrence forms in tailings, with tailings mineralogy information and heavy metal properties as input variables and the percentages of seven occurrence forms as output variables. The optimum GBRT model achieved excellent performance, with R values greater than 0.92 recorded on the testing set for all seven occurrence forms. The feature importance analysis showed that electronegativity was the most critical variable across all occurrence forms, with an average feature importance of 0.442, followed closely by atomic number, which had an average feature importance of 0.211. Overall, this study proposes a reliable and efficient GBRT prediction model for heavy metal occurrence forms, providing new insights into the effects of tailings mineralogy on heavy metal occurrence forms. This approach can be applied to contamination analysis and the safe and efficient use of heavy metals.

Appraising the crop health response to water stress from enhanced crop and soil water estimates using SAR data and machine learning approaches

ABSTRACT Precise information on soil moisture (SM) and crop water dynamics is essential for hydrological and agricultural applications. The SM measurements from SAR data are influenced by varying degree of soil-water binding and thus affect the reliable assessment of vegetation dynamics. Hence, in this present study, the six SAR-based SM descriptors capable of representing bound and free water in the soil and the corresponding in-situ SM measurements were used to train the four state-of-the-art machine learning regression (MLR) approaches, namely, random forest regression (RFR), gradient boosting regression Tree (GBRT), support vector regression (SVR), and gaussian processes regression (GPR) to retrieve enhanced surface soil moisture (SSM enh ) estimates. The SSM enh results were used in the Water Cloud Model (WCM) to retrieve the PWC (plant water content). Lastly, the crop health schema (CHS) was proposed considering SSM enh and PWC, to examine the health dynamics of the cotton and sorghum crops. The results demonstrated that the GBRT model (R2 = 0.91, RMSE = 0.004 m3m − 3, MAE = 0.021) outperformed the other ML models in retrieving SM. The PWC (R2 = 0.82, RMSE = 0.014, MAE = 0.019) from WCM and CHS schema (F1 score = 0.86, Kappa = 0.79) have shown good statistical agreement with the field observations. The present study demonstrated the scope of SSM enh product in investigating vegetation health response to water stress from single-date dual-pol SAR imagery.

Estimation of forest canopy closure in northwest Yunnan based on multi-source remote sensing data colla-boration.

Forest canopy closure (FCC) is an important parameter to evaluate forest resources and biodiversity. Using multi-source remote sensing collaborative means to achieve regional forest canopy closure inversion with low cost and high-precision is a research hotspot. Taking ICESat-2/ATLAS data as the main information source and combined with data of 54 measured plots, we estimated FCC value by the Bayesian optimization (BO) algorithm improved random forest (RF), K-nearest neighbor (KNN), and gradient boosting regression tree (GBRT) model at footprint-scale. Combined with multi-source remote sensing image Sentinel-1/2 and terrain factors, we estimated the regional-scale FCC value of Shangri-La in the northwest Yunnan based on deep neural network (DNN) optimized by BO algorithm. The results showed that six characteristic parameters (percentage of tree canopy, standard deviation of relative height of photons at the top of the canopy, minimum canopy height, difference between 98% canopy height and median canopy height in the segment, number of top canopy photons, apparent surface reflectance) out of the 50 parameters that were extracted from ATLAS lidar footprint had higher contribution rate after RF characteristic variable optimization, which could be used as model variable for footprint-scale remote sensing estimation. Among BO-RF, BO-KNN, and BO-GBRT models, the FCC results estimated by the BO-GBRT model were the best at footprint-scale. The coefficient of determination (R2) was 0.65, the root mean square error (RMSE) was 0.10, the mean absolute residual (RS) was 0.079, and the prediction accuracy (P) was 0.792 for leave-one-out cross validation. It could be used as the FCC estimation model of 74808 ATLAS footprints for forest in the study area. We used the ATLAS footprint-scale FCC value of forest as the large sample data of the regional-scale BO-DNN model and combined with multi-source remote sensing factors to estimate FCC in the study area, the accuracy of the 10-fold cross-validation BO-DNN model was R2=0.47, RMSE=0.22, P=0.558. The mean values of FCC in the study area estimated by BO-DNN model and ordinary Kriging (OK) interpolation were 0.46 and 0.52, respectively, and the values mainly distributed in 0.3-0.6, accounting for 77.8% and 81.4%, respectively. The FCC efficiency obtained directly by the OK interpolation method was higher (R2=0.26), but the prediction accuracy was significantly lower than the BO-DNN model (R2=0.49). The FCC high value was distributed from northwest to southeast in the study area, and the northern and southeastern regions were the main distribution areas of high and low FCC values, respectively. It had certain advantages to estimate mountain area FCC based on ICESat-2/ATLAS high-density footprint, and the estimation results of small sample data at footprint-scale could be used as large sample data of deep learning model at region-scale, which would provide a reference for the low-cost and high-precision to FCC estimation on the footprint-scale up to the extrapolated regional-scale.

Multiple machine learning models in estimating viscosity of crude oil: Comparisons and optimization for reservoir simulation

Development of predictive models for estimation of reservoir fluids properties as function of temperature, fluid composition, and pressure would be essential for simulation of oil reservoirs. The measurement of the viscosity of heavy crude oil is an essential part of petroleum science which can be done by experimental methods, however computational methods can be integrated to the experimental methods to reduce the effort and save time of measurements. For the purpose of predicting the heavy-oil viscosity, this work applies a number of different models to the dataset that is currently available for correlating viscosity to input parameters. The Gradient Boosting Regression Tree (GBR), the Support Vector Machine (SVM), and the Stochastic Gradient Decent (SGD) are the models that have been utilized in this investigation, and the genetic algorithm (GA) has been applied in order to optimize the hyper-parameters of these models. The RMSE error rates for the completed versions of the GBR, SGD, and SVM models are, in descending order, 26.6, 29.8, and 30.5. For the purposes of this research, the GBR model was determined to be the most suitable option among those available for estimation of crude oil viscosity.

Prediction of higher heating value of coal based on gradient boosting regression tree model

Higher heating value, also known as the coal calorific value, is an important indicator of coal quality. Nevertheless, traditional experimental determination of higher heating value is expensive, tedious, and time-consuming. Some regression models have been commonly used to solve this problem. However, the prediction accuracy still needs to be improved. Accordingly, based on the correlation between proximate analysis of coal and higher heating value, a gradient boosting regression tree model was established in this study to predict the higher heating value as a means to a less intensive procedure to estimate coal heating values. This approach used data from the U.S. Geological Survey Coal Quality database and Kentucky Coal Database to evaluate coal quality. The hyperparameters of the regression tree model were tuned by cross-validation. A comparative study was conducted between the model established in this paper and the existing methods. Experimental results indicate that the coefficient of determination of the gradient boosting regression tree model of two databases achieves 0.9862 and 0.9541, respectively. Thus, the prediction capability of the gradient boosting regression tree model outperforms the existing methods.

Analyzing of metal organic frameworks performance in CH4 adsorption using machine learning techniques: A GBRT model based on small training dataset

Metal organic frameworks (MOFs) have already been widely used in the field of gas adsorption and separation. The combination of various organic ligands and metal ions results in large MOF libraries with different topological structures. However, such large structural databases make it difficult to screen out MOF materials with high adsorption performance towards a given gas to establish statistical pictures through experiments or molecular simulations. In this work, we studied and built machine learning (ML) models based on the hypothetical MOF (hMOF) database. A gradient-boosted regression trees (GBRT) model with 1000 samples can predict the methane adsorption capacity of more than 130,000 hMOFs with high accuracy. A fast high-throughput screening on the entire database was performed using the constructed regression model, which extracts several molecular fragments from the top 1% of materials with the largest methane adsorption capacity. Careful analysis of the chemical structure of these fragments indicates that MOF materials containing cyano, hydroxyl, carboxyl and/or aromatic rings may possess high adsorption performance. The electrostatic potential distribution of these fragments was obtained at the ab initio computational level and was highlighted to be another important descriptor. We further extended the GBRT predictor to diverse MOF libraries including tobacco MOF (tobmof) and CoRE MOF. The molecular fragments extracted from the ML-based screening on the real CoRE MOF database as well as the corresponding electrostatic properties were compared with the results of hMOFs, which provides new insight into the design of MOF materials.