Landslides pose a severe threat to people, buildings, and infrastructure. The rugged terrain of the Chattogram Hill Tract region in southeastern Bangladesh frequently experiences landslides, particularly during rainy seasons. This study provides a comparative analysis of innovative machine learning (ML) algorithms used for the purpose of landslide susceptibility (LS) mapping for the Khagrachari district of Bangladesh. The dataset for this study comprises 15 landslide conditioning factors and 127 landslide inventory points. The landslide inventory points included 71 landslide and 56 non-landslide points. Then, the data were split randomly into training data (70%) and testing data (30%). Three ML algorithms, namely random forest (RF), boosted regression trees (BRT), and k-nearest neighbor (KNN), were utilized to evaluate the LS zone. The models were validated using the area under the curve (AUC), overall accuracy, precision, and recall. Based on the AUC value, the BRT model demonstrated the highest performance with a value of 0.95, while the AUC values for RF and KNN were 0.91 and 0.86, respectively. Besides, overall accuracy, precision, and recall values (0.82, 0.81, and 0.86) also indicated BRT as the most effective model. The results showed that maximum rainfall and elevation were the most influential factors for both BRT and RF models. This research provides valuable insight into understanding the LS areas in Khagrachari, aiding in informed decision-making regarding landslide-related concerns in the region, and can be applied to the broader scale to develop effective planning and mitigation strategies for comprehensive disaster management and natural hazard response.
Keywords: Boosted regression trees; K-nearest neighbor; Khagrachari; Landslide susceptibility; Machine learning; Random forest.
© 2024. The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.