Landslide susceptibility assessment using random forest classifier in machine learning in the Rwenzori Region

Abstract

Rwenzori Region is found in the Mid-Western part of Uganda and it shares the Ugandan Boarder with the Democratic Republic of Congo. This area happens to fall within the Rift Valley Corridor which is characterized by steep mountainous terrain, intense rainfall, and active tectonic activity, making it highly susceptible to landslides. In recent years, landslides triggered by heavy rainfall have caused significant damage to infrastructure, agricultural lands, and human settlements causing loss of life and economic losses. This study applies Random Forest (RF) model in Machine Learning to assess landslide susceptibility in the region. A landslide inventory map consisting of 79 landslides was prepared using field surveys. Eight causative factors—slope, soil, population density, land use, elevation, rainfall, and distance to streams and distance to roads —were analyzed. The landslide locations were divided into 60% training and 20% validation and 20% testing datasets. The resulting susceptibility maps were validated using Area under the Curve (AUC) method. The RF model showed a predictive accuracy of 91.5%. The final susceptibility map can aid in land-use planning and disaster risk reduction in the Rwenzori Region.