Proposing an ensemble machine learning based drought vulnerability index using M5P, dagging, random sub-space and rotation forest models

Abstract Drought is one of the major barriers to socio-economic development a region. To manage and reduce impact drought, drought vulnerability modelling important. The use an ensemble machine learning technique i.e. M5P, M5P -Dagging, M5P-Random SubSpace (RSS) M5P-rotation forest (RTF) assess maps (DVMs) for state Odisha in India was proposed first time. A total 248 drought-prone villages (samples) 53 indicators (DVIs) under exposure (28), sensitivity (15) adaptive capacity (10) were used produce DVMs. Out samples, 70% training models 30% validating models. Finally, DVMs authenticated by area curve (AUC) receiver operating characteristics, precision, mean-absolute-error, root-mean-square-error, K-index Friedman Wilcoxon rank test. Nearly 37.9% research region exhibited very high drought. All had capability model vulnerability. As per test, significant differences occurred among output accuracy base classifier improved after with RSS RTF meta classifiers but reduced Dagging. According validation statistics, M5P-RFT achieved highest AUC 0.901. prepared would help planners decision-makers formulate strategies reducing damage