Research Information System
Expert Systems with Applications, vol.297, 2026 (SCI-Expanded)
Landslides, highly destructive natural hazards, threaten mountainous regions impacted by rainfall, seismic activity, and human actions. Machine learning (ML) techniques for landslide susceptibility mapping (LSM) often face challenges like interpretability, overfitting, and handling high-dimensional data. This study presents Multi Hive Artificial Bee Colony Programming (MHABCP), a novel symbolic regression framework merging swarm intelligence and multi-tree programming to create interpretable, robust LSM models. A key feature is the integration of the Relative Outlier Cluster Factor method for outlier detection, enhancing data quality and model stability. Tested against Multi Gene Genetic Programming (MGGP) using a dataset from Varto, eastern Turkey, with 18 environmental and topographic factors, MHABCP achieved 90.57 % test accuracy and a 90.51 % F1-score, surpassing MGGP in all metrics while remaining interpretable. MHABCP also showed consistency across 100 runs and better classified landslide-prone areas, offering a scalable, explainable solution for disaster risk reduction and geospatial planning.