An ML-Based System for the Early Detection of Earth Slope Failures Using IoT Sensing Technology

Landslides pose significant risks to life and property, often triggered by heavy rainfall, snowmelt, earthquakes, or volcanic activity, and exacerbated by deforestation, mining, and poor construction practices. Slopes made of soft or heavily weathered materials, such as clay or shale, are especially prone to collapse. With these events becoming more frequent, robust early-warning systems are increasingly essential. As Machine Learning (ML) and the Internet of Things (IoT) are increasingly becoming popular predictive tools in infrastructure maintenance, this paper introduces an ML-based system for the early detection of earth slope failures using IoT and hydrological sensors. The developed system enables preventative interventions and delivers crucial information required for slope maintenance by monitoring slope movement in real time and predicting imminent failures. In this paper, various ML algorithms were trained, including XGBoost, Support Vector Machine (SVM), Random Forest (RF), and Long Short-Term Memory (LSTM), with a dataset collected from a prototype slope model constructed in a laboratory-scale, incorporated with an artificial rainfall system to replicate the effects of rainfall infiltration under controlled conditions. The ML models were trained using both IoT (e.g., pitch and roll) and hydrological (e.g., moisture and suction) sensory time-series data in a supervised learning environment, whereas the slope movement class serves as the class label. The proposed system was evaluated experimentally for classification accuracy using various performance metrics such as precision, accuracy, recall, and F1-score. Based on preliminary evaluation results, it was observed that our system can achieve acceptable predictions of slope movements. The RF model achieved the highest accuracy of 71.77%, closely followed by XGBoost at 71.75%, while LSTM reached 71.30%, and SVM performed the lowest with 66.22%.