Landslide Classification and Prediction of Debris Flow Using Machine Learning Models

Abstract

Landslides are unpredictable and destructive calamities that motivated the development of early warning systems to alert people and save their lives and property. A novel two-layer architecture is proposed to classify and predict landslides and types. The research is carried out in the Idukki district, Kerala, India, covering an area of 4358 km2. Layer-1 uses topographical, hydrogeological, and meteorological features collected from various sources. The binary classifier in this layer classifies and predicts the probability of landslide occurrence. Layer-2 classifies landslides into debris flow, rockfall, and surficial slides and predicts debris flow through a multi-class classifier. To achieve better prediction accuracy, the landslide features that significantly influence the landslide occurrence of layer-1 are combined with layer-2. The Synthetic Minority Oversampling Technique is used to balance the database of minority categories. Classification and prediction are achieved using machine learning models such as the support vector machine, decision tree, random forest, and extreme gradient-boosting, and their performances are compared. The feature score for every feature of the model is calculated using the feature importance technique. The machine learning models are validated using the Area Under the Receiver Operating Characteristics curve and F1-score. On comparison, Extreme Gradient-Boost is found to perform well. Based on the prediction probability, the debris flow is classified into high-risk, medium-risk, and low-risk regions. This favored in identifying vulnerable areas and appropriate mitigation measures to reduce life loss and damage to infrastructure.