Landslide susceptibility mapping using artificial intelligence models: a case study in the Himalayas

Landslides, whether natural or anthropogenic, pose significant threats to ecosystems and human lives, necessitating robust assessment methodologies. This study presents a pioneering approach by integrating deep learning (DL) and machine learning (ML) frameworks for landslide susceptibility mapping (LSM) in the Alpuri Valley, Himalayas, Pakistan. To the best of our knowledge, this is the first application of advanced DL and ML techniques in this region. The research introduces novel data representation algorithms to develop a hybrid landslide susceptibility map, representing a unique methodological advancement in LSM studies. The research examined twelve landslide-influencing factors, ensuring their suitability through multicollinearity diagnostics using tolerance, variation inflation factor, and Pearson’s correlation coefficient. A total of 162 landslide sites were randomly split into training (70%) and testing (30%) datasets. The novel hybrid support vector machine (SVM) and random forest (RF) model demonstrated remarkable predictive performance, achieving an AUROC value of 0.90 and robust results across multiple metrics, including an accuracy of 0.79, precision of 0.81, recall of 0.89, F-measure of 0.84, Matthew’s correlation coefficient of 0.43, mean squared error of 0.24, and root mean squared error of 0.48. This study represents a significant step forward in landslide susceptibility mapping by applying advanced computational models and introducing innovative hybrid techniques. The susceptibility maps generated provide a vital foundation for sustainable land use planning, infrastructure development, and disaster risk reduction, particularly in the context of regions vulnerable to landslide hazards. By advancing both the methodology and application of LSM, this research establishes a benchmark for future studies in the Himalayas and other similar terrains.