Federated Learning-Based Approach for Landslide Prediction – the Case Study in Taiwan

This study investigates the application of federated learning to regional landslide prediction in Taiwan. Rainfall and terrain data are collected and partitioned by area and year to create multiple clients, each with a distinct local dataset. Convolutional Neural Networks (CNNs) are trained independently on each client, and model weights are then aggregated on a central server. This decentralized approach addresses issues of data fragmentation and scarcity, commonly seen in landslide datasets. Experimental results demonstrate that the proposed method improves model generalization and prediction accuracy across different geographic and temporal domains, offering a scalable solution for early warning systems in data-limited regions.