Mapping landslide susceptibility at the Three Gorges Reservoir, China, using gradient boosting decision tree,random forest and information value models

This work was to generate landslide susceptibility maps for the Three Gorges Reservoir(TGR) area, China by using different machine learning models. Three advanced machine learning methods, namely, gradient boosting decision tree(GBDT), random forest(RF) and information value(InV) models, were used, and the performances were assessed and compared. In total, 202 landslides were mapped by using a series of field surveys, aerial photographs, and reviews of historical and bibliographical data. Nine causative factors were then considered in landslide susceptibility map generation by using the GBDT, RF and InV models. All of the maps of the causative factors were resampled to a resolution of 28.5 m. Of the 486289 pixels in the area,28526 pixels were landslide pixels, and 457763 pixels were non-landslide pixels. Finally, landslide susceptibility maps were generated by using the three machine learning models, and their performances were assessed through receiver operating characteristic(ROC) curves, the sensitivity, specificity,overall accuracy(OA), and kappa coefficient(KAPPA). The results showed that the GBDT, RF and In V models in overall produced reasonable accurate landslide susceptibility maps. Among these three methods, the GBDT method outperforms the other two machine learning methods, which can provide strong technical support for producing landslide susceptibility maps in TGR.

Determination of the Pile Drivability Using Random Forest Optimized by Particle Swarm Optimization and Bayesian Optimizer

Driven piles are used in many geological environments as a practical and convenient structural component.Hence,the determination of the drivability of piles is actually of great importance in complex geotechnical applications.Conventional methods of predicting pile drivability often rely on simplified physicalmodels or empirical formulas,whichmay lack accuracy or applicability in complex geological conditions.Therefore,this study presents a practical machine learning approach,namely a Random Forest(RF)optimized by Bayesian Optimization(BO)and Particle Swarm Optimization(PSO),which not only enhances prediction accuracy but also better adapts to varying geological environments to predict the drivability parameters of piles(i.e.,maximumcompressive stress,maximum tensile stress,and blow per foot).In addition,support vector regression,extreme gradient boosting,k nearest neighbor,and decision tree are also used and applied for comparison purposes.In order to train and test these models,among the 4072 datasets collected with 17model inputs,3258 datasets were randomly selected for training,and the remaining 814 datasets were used for model testing.Lastly,the results of these models were compared and evaluated using two performance indices,i.e.,the root mean square error(RMSE)and the coefficient of determination(R2).The results indicate that the optimized RF model achieved lower RMSE than other prediction models in predicting the three parameters,specifically 0.044,0.438,and 0.146;and higher R^(2) values than other implemented techniques,specifically 0.966,0.884,and 0.977.In addition,the sensitivity and uncertainty of the optimized RF model were analyzed using Sobol sensitivity analysis and Monte Carlo(MC)simulation.It can be concluded that the optimized RF model could be used to predict the performance of the pile,and it may provide a useful reference for solving some problems under similar engineering conditions.

Predicting Surface Urban Heat Island in Meihekou City, China: A Combination Method of Monte Carlo and Random Forest

Given the rapid urbanization worldwide, Urban Heat Island(UHI) effect has been a severe issue limiting urban sustainability in both large and small cities. In order to study the spatial pattern of Surface urban heat island(SUHI) in China’s Meihekou City, a combination method of Monte Carlo and Random Forest Regression(MC-RFR) is developed to construct the relationship between landscape pattern indices and Land Surface Temperature(LST). In this method, Monte Carlo acceptance-rejection sampling was added to the bootstrap layer of RFR to ensure the sensitivity of RFR to outliners of SUHI effect. The SHUI in 2030 was predicted by using this MC-RFR and the modeled future landscape pattern by Cellular Automata and Markov combination model(CA-Markov). Results reveal that forestland can greatly alleviate the impact of SUHI effect, while reasonable construction of urban land can also slow down the rising trend of SUHI. MC-RFR performs better for characterizing the relationship between landscape pattern and LST than single RFR or Linear Regression model. By 2030, the overall SUHI effect of Meihekou will be greatly enhanced, and the center of urban development will gradually shift to the central and western regions of the city. We suggest that urban designer and managers should concentrate vegetation and disperse built-up land to weaken the SUHI in the construction of new urban areas for its sustainability.

Desertification status mapping in MuttumaWatershed by using Random Forest Model

Potential of the Random Forest Model on mapping of different desertification processes was studied in Muttuma watershed of mid-Murrumbidgee river region of New South Wales,Australia.Desertification vulnerability index was developed using climate,terrain,vegetation,soil and land quality indices to identify environmentally sensitive areas for desertification.Random Forest Model(RFM)was used to predict the different desertification processes such as soil erosion,salinization and waterlogging in the watershed and the information needed to train classification algorithms was obtained from satellite imagery interpretation and ground truth data.Climatic factors(evaporation,rainfall,temperature),terrain factors(aspect,slope,slope length,steepness,and wetness index),soil properties(pH,organic carbon,clay and sand content)and vulnerability indices were used as an explanatory variable.Classification accuracy and kappa index were calculated for training and testing datasets.We recorded an overall accuracy rate of 87.7%and 72.1%for training and testing sites,respectively.We found larger discrepancies between overall accuracy rate and kappa index for testing datasets(72.2%and 27.5%,respectively)suggesting that all the classes are not predicted well.The prediction of soil erosion and no desertification process was good and poor for salinization and water-logging process.Overall,the results observed give a new idea of using the knowledge of desertification process in training areas that can be used to predict the desertification processes at unvisited areas.

Modeling the Spatial Distribution of Soil Heavy Metals Using Random Forest Model—A Case Study of Nairobi and Thirirka Rivers’ Confluence

Modeling the spatial distribution of soil heavy metals is important in determining the safety of contaminated soils for agricultural use. This study utilized 60 topsoil samples (0 - 30 cm), multispectral images (Sentinel-2), spectral indices, and ancillary data to model the spatial distribution of heavy metals in the soils along the Nairobi River. The model was generated using the Random Forest package in R. Using R2 to assess the prediction accuracy, the Random Forest model generated satisfactory results for all the elements. It also ranked the variables in order of their importance in the overall prediction. Spectral indices were the most important variables within the rankings. From the predicted topsoil maps, there were high concentrations of Cadmium on the easterly end of the river. Cadmium is an impurity in detergents, and this section is in close proximity to the Nairobi water sewerage plant, which could be a direct source of Cadmium. Some farms had Zinc levels which were above the World Health Organization recommended limit. The Random Forest model performed satisfactorily. However, the predictions can be improved further if the spatial resolutions of the various variables are increased and through the addition of more predictor variables.

Identification of Mixtures of Two Types of Body Fluids Using the Multiplex Methylation System and Random Forest Models

Objective Body fluid mixtures are complex biological samples that frequently occur in crime scenes,and can provide important clues for criminal case analysis.DNA methylation assay has been applied in the identification of human body fluids,and has exhibited excellent performance in predicting single-source body fluids.The present study aims to develop a methylation SNaPshot multiplex system for body fluid identification,and accurately predict the mixture samples.In addition,the value of DNA methylation in the prediction of body fluid mixtures was further explored.Methods In the present study,420 samples of body fluid mixtures and 250 samples of single body fluids were tested using an optimized multiplex methylation system.Each kind of body fluid sample presented the specific methylation profiles of the 10 markers.Results Significant differences in methylation levels were observed between the mixtures and single body fluids.For all kinds of mixtures,the Spearman’s correlation analysis revealed a significantly strong correlation between the methylation levels and component proportions(1:20,1:10,1:5,1:1,5:1,10:1 and 20:1).Two random forest classification models were trained for the prediction of mixture types and the prediction of the mixture proportion of 2 components,based on the methylation levels of 10 markers.For the mixture prediction,Model-1 presented outstanding prediction accuracy,which reached up to 99.3%in 427 training samples,and had a remarkable accuracy of 100%in 243 independent test samples.For the mixture proportion prediction,Model-2 demonstrated an excellent accuracy of 98.8%in 252 training samples,and 98.2%in 168 independent test samples.The total prediction accuracy reached 99.3%for body fluid mixtures and 98.6%for the mixture proportions.Conclusion These results indicate the excellent capability and powerful value of the multiplex methylation system in the identification of forensic body fluid mixtures.

Linear and Nonlinear Trading Models with Gradient Boosted Random Forests and Application to Singapore Stock Market

This paper presents new trading models for the stock market and test whether they are able to consistently generate excess returns from the Singapore Exchange (SGX). Instead of conventional ways of modeling stock prices, we construct models which relate the market indicators to a trading decision directly. Furthermore, unlike a reversal trading system or a binary system of buy and sell, we allow three modes of trades, namely, buy, sell or stand by, and the stand-by case is important as it caters to the market conditions where a model does not produce a strong signal of buy or sell. Linear trading models are firstly developed with the scoring technique which weights higher on successful indicators, as well as with the Least Squares technique which tries to match the past perfect trades with its weights. The linear models are then made adaptive by using the forgetting factor to address market changes. Because stock markets could be highly nonlinear sometimes, the Random Forest is adopted as a nonlinear trading model, and improved with Gradient Boosting to form a new technique—Gradient Boosted Random Forest. All the models are trained and evaluated on nine stocks and one index, and statistical tests such as randomness, linear and nonlinear correlations are conducted on the data to check the statistical significance of the inputs and their relation with the output before a model is trained. Our empirical results show that the proposed trading methods are able to generate excess returns compared with the buy-and-hold strategy.

A Data-Driven Car-Following Model Based on the Random Forest

The car-following models are the research basis of traffic flow theory and microscopic traffic simulation. Among the previous work, the theory-driven models are dominant, while the data-driven ones are relatively rare. In recent years, the related technologies of Intelligent Transportation System (ITS) re- presented by the Vehicles to Everything (V2X) technology have been developing rapidly. Utilizing the related technologies of ITS, the large-scale vehicle microscopic trajectory data with high quality can be acquired, which provides the research foundation for modeling the car-following behavior based on the data-driven methods. According to this point, a data-driven car-following model based on the Random Forest (RF) method was constructed in this work, and the Next Generation Simulation (NGSIM) dataset was used to calibrate and train the constructed model. The Artificial Neural Network (ANN) model, GM model, and Full Velocity Difference (FVD) model are em- ployed to comparatively verify the proposed model. The research results suggest that the model proposed in this work can accurately describe the car- following behavior with better performance under multiple performance indicators.

基于PSO-RF的妊娠母猪日饲喂量预测算法研究

日饲喂量对妊娠期母猪繁殖性能具有较大影响,对于保障母猪健康、胎儿生长发育具有重要意义。为了精准控制日饲喂量,针对粒子群算法(PSO)各阶段搜索能力不均衡的问题,引入一种非线性递减惯性权重策略对PSO算法进行改进,并用改进的粒子群优化随机森林回归算法(PSO-RF)精确预测妊娠母猪日饲喂量,精准控制智能饲喂器的饲料投放。该算法融合随机森林的高准确性和粒子群算法的参数寻优能力强的特性,通过优化决策树的数量和最大深度来提升预测性能。结果表明,PSO-RF算法取得的决定系数R^(2)值达到0.9814,相较于RF算法、SVM支持向量机和BP神经网络分别提升了1.19%、2.30%和3.25%。PSO-RF算法在预测妊娠母猪日饲喂量方面具有更高的精准度,有助于提高养猪场管理的智能化水平,降低生产成本,提升养猪场养殖效益,具有一定实际应用价值。

Design and analysis of traffic incident detection based on random forest

In order to avoid the noise and over fitting and further improve the limited classification performance of the real decision tree, a traffic incident detection method based on the random forest algorithm is presented. From the perspective of classification strength and correlation, three experiments are performed to investigate the potential application of random forest to traffic incident detection： comparison with a different number of decision trees; comparison with different decision trees; comparison with the neural network. The real traffic data of the 1-880 database is used in the experiments. The detection performance is evaluated by the common criteria including the detection rate, the false alarm rate, the mean time to detection, the classification rate and the area under the curve of the receiver operating characteristic （ROC）. The experimental results indicate that the model based on random forest can improve the decision rate, reduce the testing time, and obtain a higher classification rate. Meanwhile, it is competitive compared with multi-layer feed forward neural networks （MLF）.

基于K-means SMOTE和IDBO-RF岩爆烈度等级预测模型

为解决岩爆数据集不均衡和模型参数寻优困难等问题,提出1种基于K-means SMOTE与改进蜣螂算法优化随机森林(random forest,RF)的预测模型。首先,分析岩爆发生机理构建指标体系;其次,使用K-means SMOTE算法对岩爆数据集进行均衡化处理,采用Robust标准化消除量纲;最后,引入Tent混沌映射和非线性递减策略组合改进蜣螂优化(improved dung beetle optimizer,IDBO)算法,寻优RF超参数,建立岩爆烈度等级预测模型(IDBO-RF)并与其他模型对比验证其有效性。研究结果表明:数据均衡处理后,各模型准确率提高10.85%~16.02%;设计的IDBO-RF预测模型平均准确率约为94.37%,较RF、GWO-RF、DBO-RF模型分别提高约7.76百分点、1.69百分点、1.11百分点;IDBO-RF预测模型准确率最高约为96.43%,优于RF、GWO-RF、DBO-RF模型。研究结果可为解决岩爆预测问题提供一定参考。

基于RF-GWO的水利工程地质渗透系数智能反演分析

地质渗透系数是准确分析水利工程渗流的关键参数。针对传统反演方法计算效率低、精度差的问题,采用有限元正演模型和正交试验设计构建渗透系数反演样本集,建立了基于随机森林(RF)算法的渗流计算代理模型;在此基础上,引入灰狼优化(GWO)算法,提出了基于RF-GWO的渗透系数智能反演方法,并以Z抽水蓄能电站为研究案例进行了验证。结果表明:RF模型对各钻孔水位预测结果均接近实测值,性能优于CART和BP模型;GWO可搜寻到地质最佳渗透系数,钻孔水位反演结果合理,相对误差最大为0.42%,精度满足工程要求,计算的天然渗流场分布形态也符合一般山体渗流场分布规律。建立的反演模型能够快速准确地推断工程区地层渗透系数,具有实际工程应用价值。

基于LSTM-RF的电动钻机绞车齿轮箱故障诊断

针对提高石油电动钻机绞车齿轮箱故障诊断的准确性和效率,提出了一种基于长短期记忆网络(long short-term memory,LSTM)和随机森林(random forest,RF)融合模型。首先,运用LSTM能够从大规模数据中学习复杂特征,将这些特征作为随机森林的输入。然后,通过随机森林处理非线性和高维数据以及对特征的分类,以实现对齿轮不同故障状态的识别。最后,利用电动钻机绞车齿轮箱运行过程中的实时数据,建立了一个包含多种齿轮故障类型的综合数据集。试验结果表明,LSTM齿轮故障诊断准确率为94.67%,RF齿轮故障诊断准确率为94.34%,支持向量机齿轮故障诊断准确率为82.00%,K近邻齿轮故障诊断准确率88.33%,而融合模型LSTM-RF在齿轮故障诊断准确率方面达到了98.33%,克服了单一模型的局限性,提高了诊断准确性。研究表明了融合模型具有更优的电动钻机绞车齿轮箱故障诊断能力。

基于RF-SFLA-SVM的装配式建筑高空作业工人不安全行为预警

为有效预警装配式建筑高空作业工人不安全行为的发生趋势或状态,增强对装配式建筑工人不安全行为(PBWUBs)的管控,采用随机森林(RF)-混合蛙跳算法(SFLA)-支持向量机(SVM)模型,开展工人不安全行为预警研究。首先,采用SHEL模型分析处于高空作业危险中的PBWUBs的影响因素,并通过RF确定关键预警指标;然后,采用SFLA对SVM的参数进行寻优改进;最后,利用RF-SFLA-SVM预警高空作业PBWUBs,提出应对措施,并与其他预警模型对比。研究结果表明:基于RF-SFLA-SVM预警高空作业PBWUBs,准确率最高,为91.67%,与其他模型的预警性能相比,最高提升14%。研究结果可为高空作业PBWUBs的防控提供参考。

Mixed-effects modeling for tree height prediction models of Oriental beech in the Hyrcanian forests

Height–diameter relationships are essential elements of forest assessment and modeling efforts.In this work,two linear and eighteen nonlinear height–diameter equations were evaluated to find a local model for Oriental beech（Fagus orientalis Lipsky） in the Hyrcanian Forest in Iran.The predictive performance of these models was first assessed by different evaluation criteria： adjusted R^2（R^2_（adj））,root mean square error（RMSE）,relative RMSE（%RMSE）,bias,and relative bias（%bias） criteria.The best model was selected for use as the base mixed-effects model.Random parameters for test plots were estimated with different tree selection options.Results show that the Chapman–Richards model had better predictive ability in terms of adj R^2（0.81）,RMSE（3.7 m）,%RMSE（12.9）,bias（0.8）,%Bias（2.79） than the other models.Furthermore,the calibration response,based on a selection of four trees from the sample plots,resulted in a reduction percentage for bias and RMSE of about 1.6–2.7%.Our results indicate that the calibrated model produced the most accurate results.

A rapid classification method of aluminum alloy based on laser-induced breakdown spectroscopy and random forest algorithm

As an important non-ferrous metal structural material most used in industry and production,aluminum(Al) alloy shows its great value in the national economy and industrial manufacturing.How to classify Al alloy rapidly and accurately is a significant, popular and meaningful task.Classification methods based on laser-induced breakdown spectroscopy(LIBS) have been reported in recent years. Although LIBS is an advanced detection technology, it is necessary to combine it with some algorithm to reach the goal of rapid and accurate classification. As an important machine learning method, the random forest(RF) algorithm plays a great role in pattern recognition and material classification. This paper introduces a rapid classification method of Al alloy based on LIBS and the RF algorithm. The results show that the best accuracy that can be reached using this method to classify Al alloy samples is 98.59%, the average of which is 98.45%. It also reveals through the relationship laws that the accuracy varies with the number of trees in the RF and the size of the training sample set in the RF. According to the laws, researchers can find out the optimized parameters in the RF algorithm in order to achieve,as expected, a good result. These results prove that LIBS with the RF algorithm can exactly classify Al alloy effectively, precisely and rapidly with high accuracy, which obviously has significant practical value.

CHAID-RF:基于CHAID决策树的集成学习方法

针对卡方自动交互诊断(CHAID)决策树易过拟合的问题,提出CHAID随机森林方法(CHAID Random Forest,CHAID-RF)。该方法采用随机采样、随机选择特征以及集成的策略,将CHAID决策树作为基分类器,形成CHAID-RF。为了验证CHAID-RF的有效性,选取CART、CHAID、SVM、RF作为对比算法,以准确率、加权查准率、加权查全率、加权F值作为分类模型评价指标,以均方根误差作为回归模型评价指标,采用10个分类数据集和7个回归数据集进行验证。实验结果表明CHAID-RF可行有效。

基于RF-RNN模型的DNS隐蔽信道检测方法

为提高检测隐蔽信道的灵敏度,提出一种基于随机森林(Random Forest,RF)和循环神经网络(Recurrent Neural Network,RNN)的域名系统(Domain Name System,DNS)隐蔽信道检测方法。该方法采用域名检测作为主要手段,使用RF模型对域名进行分类,通过深度学习方法挖掘更高阶的特征表示。实验结果表明,与单一模型相比,该方法在检测准确性和健壮性方面均取得了显著提升。

Traffic flow prediction of urban road network based on LSTM-RF model

Traffic flow prediction,as the basis of signal coordination and travel time prediction,has become a research point in the field of transportation.For traffic flow prediction,researchers have proposed a variety of methods,but most of these methods only use the time domain information of traffic flow data to predict the traffic flow,ignoring the impact of spatial correlation on the prediction of target road segment flow,which leads to poor prediction accuracy.In this paper,a traffic flow prediction model called as long short time memory and random forest(LSTM-RF)was proposed based on the combination model.In the process of traffic flow prediction,the long short time memory(LSTM)model was used to extract the time sequence features of the predicted target road segment.Then,the predicted value of LSTM and the collected information of adjacent upstream and downstream sections were simultaneously used as the input features of the random forest model to analyze the spatial-temporal correlation of traffic flow,so as to obtain the final prediction results.The traffic flow data of 132 urban road sections collected by the license plate recognition system in Guiyang City were tested and verified.The results show that the method is better than the single model in prediction accuracy,and the prediction error is obviously reduced compared with the single model.

Water Quality Index Using Modified Random Forest Technique: Assessing Novel Input Features

Water quality analysis is essential to understand the ecological status of aquatic life.Conventional water quality index(WQI)assessment methods are limited to features such as water acidic or basicity(pH),dissolved oxygen(DO),biological oxygen demand(BOD),chemical oxygen demand(COD),ammoniacal nitrogen(NH3-N),and suspended solids(SS).These features are often insufficient to represent the water quality of a heavy metal–polluted river.Therefore,this paper aims to explore and analyze novel input features in order to formulate an improved WQI.In this work,prospective insights on the feasibility of alternative water quality input variables as new discriminant features are discussed.The new discriminant features are a step toward formulating adaptive water quality parameters according to the land use activities surrounding the river.The results and analysis obtained from this study have proven the possibility of predicting WQI using new input features.This work analyzes 17 new input features,namely conductivity(COND),salinity(SAL),turbidity(TUR),dissolved solids(DS),nitrate(NO3),chloride(Cl),phosphate(PO4),arsenic(As),chromium(Cr),zinc(Zn),calcium(Ca),iron(Fe),potassium(K),magnesium(Mg),sodium(Na),E.coli,and total coliform,in predicting WQI using machine learning techniques.Five regression algorithms-random forest(RF),AdaBoost,support vector regression(SVR),decision tree regression(DTR),and multilayer perception(MLP)-are applied for preliminary model selection.The results show that the RF algorithm exhibits better prediction performance,with R2 of 0.974.Then,this work proposes a modified RF by incorporating the synthetic minority oversampling technique(SMOTE)into the conventional RF method.The proposed modified RF method is shown to achieve 77.68%,74%,69%,and 71%accuracy,precision,recall,and F1-score,respectively.In addition,the sensitivity analysis is included to highlight the importance of the turbidity variable in WQI prediction.The results of sensitivity analysis highlight the importance of certain water quality variables that are not present in the conventional WQI formulation.