Random Forest Classifier-Based Landslide Susceptibility Mapping for West Bengal

Abstract

Communities in hilly and mountainous areas are seriously at risk for their safety and way of life due to recurring events of landslides. Landslides frequently occur in West Bengal, India, resulting in significant infrastructure damage and fatalities. The current research utilizes a machine learning approach based on geographic information systems to create the West Bengal region’s landslide susceptibility map (LSM). Random forest classifier (RFC) has been selected from several Blackbox models since it has been proven to be a useful tool for creating reliable map of landslide susceptibility. To conduct the analysis, an inventory was made using historical landslide data gathered from several web sources. The environmental factors considered to analyze the study area are Elevation, Slope, Aspect, Flow Accumulation and Topographic Wetness Index (TWI). Other factors also were considered, like Distance from Lineament and Fault, Distance from railway or roadway, Average Annual Rainfall, and Normal Density Vegetation Index (NDVI). The model was trained on a 80% subset of the data and then validated on the remaining 20% data. The results showed that the RFC model accurately predicted the landslide occurrences, and the confusion matrix analyses the accuracy to be 92.12%. The model’s results created the LSM for the West Bengal (WB) region. Areas with a high, moderate, and low susceptibility to landslides were indicated on the map. In this location, the model overperforms by categorizing undesirable points as landslide prone. But it indicates the requirement of further study in the district of Purulia. The created LSM can be useful for different planning in the WB as well as can be extended to other regions, offering relevant data for decision-making. The accuracy of analysis can be further increased with better resolution field data. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.