Landslide Risk Assessment Using Bayesian Network at Uttarkashi District, Uttarakhand (India)

Abstract

Landslides are a continuous disaster that jeopardizes the land and habitats due to haphazard developmental activities in the hilly regions. Machine Learning approach was used to construct landslide risk prediction models using past landslide repository. The study aims to present a powerful and capable model that is a hybrid combination of landslide domain data and Bayesian networks. Multiple causative factors that influence the landslides are also integrated to facilitate the study. The entire study is carried out in several steps that are discussed in the paper. The Bayesian network classifier exhibits good predictive capabilities with (BNTraining= 85%, BNTesting=84%) and the AUC value is (BNTraining= 0.888, BNTesting=0.928). The output labels for predictor class risk are low, moderate, and high. The results were validated and tested using different datasets. The proposed risk prediction model could be useful for government authorities and other disaster management team to make necessary decisions on time to avoid damage of settlements and infrastructure.

Keywords:Decision Support System, Factors, Landslide Risk, Landslide Data, Machine Learning.