Abstract

Landslides are devastating natural disasters, making robust early warning systems essential for mitigating their impact. This study details the development of a low-cost, low-power Internet of Things (IoT) system for real-time landslide monitoring. The system utilises prototype nodes, each integrating sensors for rainfall, soil moisture, geophones, temperature, and humidity. Data is transmitted via a Long-Range Wide-Area Network (LoRaWAN) to cloud storage, with a local SD card for backup. Data quality was rigorously validated using metrics such as Signal-to-Noise Ratio (SNR) and Mean Absolute Error (MAE). An unsupervised machine learning model, Isolation Forest, was deployed to detect anomalies from sensor data indicative of potential landslide events. The model proved effective in identifying high-risk conditions, achieving an F1-score of 0.73 for detecting critical events. While this unsupervised approach is promising for remote areas with limited baseline data, future work will involve developing supervised models using labeled field data to further improve predictive accuracy.