Comparison of machine learning algorithms combined with metaheuristic-based feature selection methods for debris flow susceptibility assessments

The selection of optimal features from large datasets for implementing machine learning algorithms has become challenging, particularly in debris flow susceptibility. Novel hybrid models, combining machine learning algorithms with feature selection methods, are proposed in this study to assess debris flow events in Diebu County, China. In total, 18 affecting factors and 195 debris flows were considered. In addition, three metaheuristic algorithms, including particle swarm optimization (PSO), grey wolf optimizer (GWO), and whale optimization algorithm (WOA), are introduced to select optimal factors for two traditional machine learning methods, namely support vector machine (SVM) and random forest (RF). Six hybrid models were built, including PSO-SVM, GWO-SVM, WOA-SVM, PSO-RF, GWO-RF, and WOA-RF. The efficiencies of the built models are assessed using the receiver operating characteristic curve and five statistical metrics. The Wilcoxon signed-rank test and boxplots are also introduced to show the statistical differences among models. The results indicated that the hybrid models obviously outperform the regular model. In addition, the number of selected factors and execution time of the hybrid models are substantially lower than those of the initial dataset. This study demonstrates the efficiency and effectiveness of metaheuristic algorithms in the pre-processing steps used for assessing debris flow susceptibility.