Integrating Feature Selection and Assisted Predictive Modelling for Enhanced Landslide Susceptibility Mapping

Addressing feature redundancy and enhancing model generalizability remain critical challenges in landslide susceptibility mapping (LSM). This study evaluated the comparative efficacy of two optimization‐based predictive frameworks, the Bayesian Assisted Predictive framework (BAYM) and the Capuchin Search Algorithm Assisted Predictive framework (CAPM), to develop high‐precision LSMs for Bijie, China. Based on a comprehensive field survey of 614 landslides, seven pivotal factors, including Slope, Aspect, Normalized Difference Vegetation Index, Elevation, Distance From River Index, Stream Power Index and Rainfall, were prioritized from 15 potential attributes using Variance Inflation Factor, Relief and Gain Ratio analyses. Subsequently, six machine learning classifiers were integrated with BAYM and CAPM for hyperparametre optimization. Performance evaluation using metrics such as Accuracy, F1‐score and Kappa statistics revealed that the Support Vector Machine optimized by BAYM, incorporating Relief‐selected features, achieved superior performance (Accuracy: 0.9846; Kappa: 0.9351). Rigorous validation on independent test sets and multiple optimization iterations confirmed the robustness of the framework against overfitting. The resulting LSM provides a high‐precision tool for hazard management and risk mitigation in landslide‐prone regions. This study contributes to the field by demonstrating that synergizing targeted feature selection with meta‐heuristic optimization significantly improves the reliability and generalizability of susceptibility models.