Landslide susceptibility mapping using hybrid machine learning classifiers: a case study of Neelum Valley, Pakistan

Abstract

The Neelum Valley in the Himalayan region of Pakistan frequently experiences landslides triggered by heavy rainfall, seismic activity, and human interventions, leading to significant loss of life and damage to infrastructure. Studying landslides in this region is essential for effective disaster management and risk assessment. Landslide susceptibility mapping (LSM) in areas like the Neelum Valley is crucial for proactive hazard mitigation and sustainable development. In this study, innovative data visualization approaches are used to create hybrid LSM, representing an innovation in LSM studies. This study aims to examine and compare the performance of various Machine Learning (ML) classifiers, including Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Categorical Boosting (CatBoost) as well as a Hybrid classifier (XGB + LGBM + CatB), for mapping landslide susceptibility. Utilizing various data sources, a total of 360 landslide locations were initially identified in the Neelum Valley to create a comprehensive landslide inventory map. These locations were then randomly split into two datasets, using a 70/30 ratio, for training and validation purposes. In the second step, a landslide factor database was developed, consisting of 14 factors related to hydrology, climate, geology, topography, and human activities. Subsequently, Pearson's correlation coefficient and the ReliefF technique were applied to rank the importance of these factors. Additionally, several performance metrics were used to evaluate the classifiers, including the area under the receiver operating characteristic curve (AUROC), accuracy, precision, recall, F-measure, Matthew's correlation coefficients, root mean square error, mean square error, Cohens Kappa and Jaccard Index. The Hybrid classifier (XGB + LGBM + CATB) achieved the highest AUC value (0.9039), indicating it was the most efficient model, followed by XGBoost, LightGBM, and CatBoost, with AUC values of 0.8920, 0.8935, and 0.8945, respectively. The study concludes that the stacking ensemble classifier shows significant potential for LSM in Neelum Valley, effectively identifying landslide-prone areas.