A Study on Performance Comparisons between KNN, Random Forest and XGBoost in Prediction of Landslide Susceptibility in Kota Kinabalu, Malaysia

One of the most natural catastrophes in Malaysia, landslides, has resulted in several fatalities, infrastructure damage and economic losses. Over time, researchers have used various methods to forecast the vulnerability to landslides. Unfortunately, the most accurate algorithm which can be used to develop a landslide susceptibility model is still lacking. Therefore, the current study aims to evaluate how well Kota Kinabalu, Sabah's landslide susceptibility, can be predicted using three different machine learning techniques: K-Nearest Neighbor (KNN), Random Forest, and Extreme Gradient Boosting (XGBoost). The research areas had 242 landslide locations, and the inventory data was arbitrarily separated into training and testing datasets in a 7/3 ratio. As prediction parameters, ten spatial databases of landslides conditioning factors were employed. The area under the curve (AUC) was utilized as the models' performance metric. With an AUC score of 87.52 %, the final analysis showed that KNN had the highest prediction accuracy, followed by Random Forest (84.34 %) and XGBoost (78.07%). According to the AUC findings, KNN, Random Forest, and XGBoost performed consistently well in forecasting landslide susceptibility. The final forecast map can be a helpful tool for urban planning and development and for aiding the authorities in creating a strategic mitigation plan.