Development of a Wilks feature importance method with improved variable rankings for supporting hydrological inference and modelling

Abstract. Feature importance has been a popular approach for machine learning models to investigate the relative significance of model predictors. In this study, we developed Wilks feature (WFI) method hydrological inference. Compared with conventional methods such as permutation (PFI) and mean decrease impurity (MDI), proposed WFI aims provide more reliable variable rankings To achieve this, measures scores based on Λ (a test statistic that can be used distinguish differences between two or groups variables) throughout an inference tree. PFI MDI methods, does not rely any performance evaluate rankings, which thus result in less biased criteria selection during tree deduction process. The was tested by simulating monthly streamflows 673 basins United States applied three interconnected irrigated watersheds located Yellow River basin, China, through concrete simulations their daily streamflows. Our results indicated could generate stable response reduction irrelevant addition, WFI-selected predictors helped random forest (RF) its optimum predictive accuracy, indicates identify informative than other measures.