Using Decision Tree J48 Based Machine Learning Algorithm for Flood Susceptibility Mapping: A Case Study in Quang Binh Province, Vietnam

Abstract

Flood is a dangerous natural hazard causing loss of life, property, and infrastructure besides affecting the economy and social life of a country. Vietnam is facing flood hazard problem recurrently mainly due to its topography, hydrology and climatic condition; and dense population. Therefore, it is necessary to develop an accurate flood susceptibility map to identify the areas likely to be affected by flooding to minimize potential damages of the flood hazard. This study deals with flood-susceptibility mapping using machine learning based on J48 decision trees algorithm for identifying in advance flood susceptible areas. In this study, we have used ten important flood affecting factors, namely Elevation, Slope, Curvature, River Density, Distance from River, Geomorphology, Land Use, Flow Accumulation, Flow Direction, and Rainfall, for the model development. Flood inventory of 318 locations was used for randomly splitting dataset in 70:30 ratio for model training (70%) and validation (30%). The model performance was evaluated using standard statistical measures. The results indicated that the decision tree (J48) model has high predictive performance (AUC = 0.97). Therefore, this model can be used to develop a reliable and accurate flood susceptibility map of an area for proper planning, flood risk management, and mitigation.