Integration of multiple optimization algorithms and coupling strategies for the enhancement of MLP-based landslide susceptibility assessment

Landslide susceptibility assessment is essential for disaster risk prevention and regional planning. This study focuses on the Chenghai–Binchuan fault zone in northwestern Yunnan, China, characterized by complex geology and frequent landslides. Fourteen influencing factors, including elevation, distance to faults, fractional vegetation cover (FVC) and land type, were selected to construct the dataset. A multi-layer per-ceptron (MLP) model was optimized using differential evolution (DE), genetic algorithm (GA), particle swarm optimization (PSO) and whale optimization algorithm (WOA), along with hybrid strategies (DE–WOA, GA–WOA, PSO–WOA). Results show that optimized models outperform the baseline MLP (accuracy 0.749; ROC-AUC 0.845), with WOA–MLP achieving 0.836 accuracy (ROC-AUC 0.861). The DE–WOA–MLP model performed best (ROC-AUC 0.877). Low and extremely low susceptibility dominate the region, while high-risk zones cluster near faults, the Jinsha River and Chenghai Lake in the southeast, forming a clear spatial gradient. These findings support targeted hazard management and future assessments.