A New Random Forest Applied to Heavy Metal Risk Assessment

摘要

As soil heavy metal pollution is increasing year by year, the risk assessmentof soil heavy metal pollution is gradually gaining attention. Soil heavy metaldatasets are usually imbalanced datasets in which most of the samples are safesamples that are not contaminated with heavy metals. Random Forest (RF) hasstrong generalization ability and is not easy to overfit. In this paper, we improvethe Bagging algorithm and simple voting method of RF. AW-RF algorithm basedon adaptive Bagging and weighted voting is proposed to improve the classificationperformance of RF on imbalanced datasets. Adaptive Bagging enables treesin RF to learn information from the positive samples, and weighted voting methodenables trees with superior performance to have higher voting weights. Experimentswere conducted using G-mean, recall and F1-score to set weights, and theresults obtained were better than RF. Risk assessment experiments were conductedusing W-RF on the heavy metal dataset from agricultural fields around Wuhan. Theexperimental results show that the RW-RF algorithm, which use recall to calculatethe classifier weights, has the best classification performance. At the end of thispaper, we optimized the hyperparameters of the RW-RF algorithm by a Bayesianoptimization algorithm. We use G-mean as the objective function to obtain the optimalhyperparameter combination within the number of iterations.