Data-driven Forecasting of Landslide Displacement Using Machine Learning and Real-time Monitoring

Abstract

Landslides pose significant risks to infrastructure and human safety, especially in areas characterized by steep terrain and unstable geological conditions. This study explores the application of machine learning (ML) models for predicting landslides in Lac Duong Town, Lam Dong Province, Vietnam, with a particular focus on the influence of rainfall and pore water pressure (PWP). The dataset comprises hourly rainfall, pore water pressure (PWP), and soil displacement records collected by an automatic monitoring system from 2020 to 2024. Three ML models, Decision Tree (DT), Random Forest (RF), and Long Short-Term Memory (LSTM), were employed to examine the relationship between cumulative rainfall, soil displacement, and landslide occurrence. The analysis identified critical cumulative rainfall thresholds (Rx3d, Rx5d, Rx7d, and Rx10d) as key indicators of landslide events. Among the models, the RF algorithm achieved the highest predictive performance, with an R² of 0.6406 and an RMSE of 0.0393, outperforming both DT and LSTM. The findings underscore the importance of cumulative rainfall in landslide forecasting and demonstrate the potential of ML, particularly RF, to enhance early warning systems. The study also proposes data-driven rainfall thresholds validated against historical landslide records. Recommendations include implementing real-time monitoring systems, refining threshold-based models, integrating landslide risk into urban planning, and including additional hydrogeological variables in future research. This work highlights the promise of ML-based approaches in improving landslide prediction and risk mitigation in vulnerable regions.

Keywords: Machine learning (ML), rainfall thresholds, early warning systems, pore water pressure, soil displacement.