Machine learning driven landslide susceptibility prediction for the Uttarkashi region of Uttarakhand in India.

A landslip is derived from nature and sometimes caused by human activities that endanger humanity and habitation. This work focuses on landslide behaviour for sustainable landslide mitigation and susceptibility analysis. The current research aims to evaluate three state-of-the-art machine learning techniques including, Random Forest (RF), Backpropagation Neural Network (BPNN), and Bayesian Network (BN), to predict the landslide susceptibility for the Uttarakashi district, Uttarakhand (India). For this purpose, a total of 554 landslide locations and eleven influencing factors were integrated to construct a landslide susceptibility map. The landslide inventory data were separated into training and testing sets. The receiver operating characteristic (ROC) with other statistical metrics, including sensitivity, specificity, precision, recall, and accuracy, was applied to compare the performance of machine learning techniques. The results indicated that the area under the curve (AUC) value for the RF model (AUC=0.89) is high. Furthermore, our findings showed that the RF model performs the best among all the other models with the highest training and testing accuracy, 96% and 86%, respectively. The final landslide susceptibility map is grouped into three classes, i.e. Low, Moderate, and High. The study outcome would provide support to disaster management officials in effective decision-making to prioritise necessary actions. [ABSTRACT FROM AUTHOR]