Sobha watershed, located in the Puruliya district of West Bengal, India, is experiencing severe soil erosion due to specific geo-environmental settings and unscientific land practices. It poses serious threats to agricultural and natural resource development, resulting in land degradation and desertification. This study attempts to identify soil erosion susceptible zones (SESZ) of the Sobha watershed by utilising remote sensing and GIS data products in different machine learning algorithms i.e., Support Vector Machine (SVM), Classification and Regression Tree (CART), Boosted Regression Tree (BRT), and Random Forest (RF)) considering sixteen soil erosion controlling factors (SECFs). In addition, the efficiency of the chosen machine learning models was evaluated using known soil erosion and non-erosion data. The results showed that elevation, drainage density (DD), and normalised difference vegetation index (NDVI) factors contribute the most to soil erosion. The ROC (receiver operating curve) AUC (area under the curve) is used to compare each model, and it was reveals that the RF model performed and predicted the best among them. However, all the models exhibit an outstanding capacity with AUC > 85% (RF = 0.97, BRT = 0.96, SVM = 0.95, and CART = 0.88). The RF model results show that the Northeastern portion of the catchment (upper part) is most vulnerable to erosion, and about 14.48% of the basin areas are under the severe erosion zone. Thereby, the findings based on machine learning algorithms and intensive field visits are utilised to assess the soil erosion risk zones, and this work will give insight into implementing suitable policies to mitigate this issue. Furthermore, the approaches utilised in this study could be useful in predicting soil erosion risk in other regions as well. In addition, the study has also recommended some appropriate policies and management approaches that would be immensely useful for the local government and policymakers in initiating strategic planning to combat soil erosion.