Integrating Geo-Environmental Factors and Ensemble Machine Learning for Landslide Susceptibility Assessment in Kodaikanal and its Environs, India

Landslide susceptibility is significant for disaster mitigation and sustainable land-use planning in geologically unstable areas, like Kodaikanal, South India. This present study examined the predictive capabilities of five machine-learning algorithms, Random Forest (RF), Gradient Boosting (GB), k-Nearest Neighbors (k-NN), Extra Trees (ET), and stacking ensemble model, along with a Frequency Ratio (FR) method, to create a landslide susceptibility map. The 70% of landslide and non-landslide points were used for training, and the remaining 30% testing to evaluate the model accuracy. The final susceptibility map was divided the area into five categories: very low, low, moderate, high, and very high. Zones with very high (7%), high (8%) were mainly located on steep slopes. The model performance was evaluated using Mean Squared Error (MSE), Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), and receiver operating characteristic (ROC) curves. Overall, the GB showed the lowest MSE, RMSE, and MAE values, indicating the highest prediction accuracy, and RF also performed strongly, ET had moderate performance, and k-NN had weaker prediction. The Stacking ensemble has higher predictive accuracy with an AUC-ROC of 0.96%. This integrated approach supports early warning system, and sustainable land management in landslide-prone areas of Kodaikanal.