<![CDATA[Landslides are among the most destructive natural hazards, causing severe casualties, economic losses, and environmental degradation. Central Sulawesi, characterized by active tectonics such as the Palu-Koro fault, is highly susceptible to landslides, as tragically demonstrated in 2018. Therefore, developing accurate landslide susceptibility maps is essential to support comprehensive landslide mitigation efforts in this region. While machine learning, particularly Random Forest (RF), has proven highly effective for landslide modeling, previous studies around Palu have often overlooked model simplification through feature selection and hyperparameter optimization. This study proposes an integrated approach combining RF with Recursive Feature Elimination (RFE) to reduce model complexity and enhance predictive accuracy. This research utilizes 498 landslide events with fifteen conditions, including topography, environment, geology, and anthropogenic influences. The RFE-RF model achieves superior classification performance, with accuracy, balanced accuracy, and F1-scores exceeding 0.81, outperforming the RF without RFE and Logistic Regression baselines. These findings underscore the urgent need to integrate feature selection methods such as RFE into landslide modeling frameworks to improve predictive accuracy. High accuracy enables government authorities and stakeholders to develop more targeted and effective mitigation priorities. Spatial analysis indicates that Donggala, Palu, and Sigi are the most critical areas requiring prioritized mitigation, with over 9% of their territories classified as highly susceptible. Feature importance analysis reveals that elevation, slope, and land cover are the most influential factors. This study suggests that mitigation efforts should focus on the hills and mountainous areas on both sides of the Palu Valley, with recommended strategies emphasizing land cover management practices, such as reforestation, to enhance slope stability and reduce landslide risk.]]>
Copyrights © 2026
