Prediction and mapping of landslide susceptibility in western mountainous areas of Xingtai City based on multiobjective optimization algorithm

Predicting landslide susceptibility with a single model often yields suboptimal accuracy in geologically complex regions. To address this limitation, we propose a novel integrated framework that synergistically combines the Information Value (IV) model, machine learning (ML) algorithms, and the Non-dominated Sorting Genetic Algorithm II (NSGA-II) for the western mountainous area of Xingtai City. Our methodology begins by employing the IV model to quantify the predictive contribution of various causative factors. Subsequently, base models are constructed by integrating IV with RF and XGBoost, respectively. The NSGA-II is then applied to concurrently optimize feature selection and model hyperparameters, culminating in four distinct coupled models. A comprehensive evaluation utilizing multiple metrics— including Accuracy, Precision, Recall, F1-score, and AUC value—revealed that the IV-RF-NSGA-II model achieved superior predictive performance, with AUC values of 0.926 on the test set and 0.918 on the training set. Furthermore, validation using the Seed Cell Area Index (SCAI) method confirmed that the susceptibility zonation map produced by this model is the most scientifically rational. Factor contribution analysis identified slope, precipitation, lithology, and peak ground acceleration (PGA) as the predominant landslide influencing factors. This study not only offers a practical tool for landslide risk management in Xingtai but also serves as a valuable reference for susceptibility assessment in other regions with analogous complex terrains.