Use of machine learning algorithms to assess flood susceptibility in the coastal area of Bangladesh.

The geographical location of Bangladesh makes it quite vulnerable to natural disasters. Flooding is one of those disasters. The coastal area of Bangladesh has higher vulnerability to flooding which is important to assess for planning and development. The study focuses on the coastal flood susceptibility through a holistic approach by combining three machine learning algorithms, namely Random Forest (RF), Extreme Gradient Boosting (XGBoost), and K Nearest Neighbor (KNN), with GIS techniques. Nine flood conditioning factors (Aspect, Slope, Elevation, Mean Sea Level, Wind, Distance from coast, Distance from river, Distance from cyclone and Geology) were considered in this study. These variables are selected by reviewing site-specific available literature and experts opinions. Relevant data were collected from several sources, and these data were analyzed in various software. The study also incorporates an analysis utilizing the best-fitted model among the three algorithms from a humanistic perspective. After obtaining results from the models, the output went under accuracy assessment to justify whether the model yielded accurate results or not. A comparative analysis to understand the suitability of these three algorithms is presented in the study, along with the delineation of five categories of flood susceptible classes of the study area. The Random Forest model was the most accurate, with an accuracy of 86.7 percent, followed by XGBoost with an accuracy of 86.3 percent and KNN with an accuracy of 85.5 percent. The Random Forest (RF) model was selected for further analysis as it yielded better accuracy incorporating the Upazila and population data. Further analysis shows that the district of Barguna, Bhola, and Patuakhali had the highest number of very high flood susceptible Upazilas among the 19 coastal districts in the study area. Around 10.63% of the study area's population, which is 4,214,084 people approximately, live in the Very High Susceptibility Zone. The results derived from the study, particularly from the humanistic perspective, are quite significant as policymakers better understand the overall situation and take necessary actions to make those areas and the communities residing in those areas more resilient and adaptive to this kind of disaster. • Assessment of Flood Susceptibility using three machine learning methods namely Random Forest (RF), Extreme Gradient Boosting (XGBoost), and K-Nearest Neighbor (KNN). • Among them Random Forest method shows better accuracy. • Distance from the coast, Distance from Cyclone Tracks cyclone, distance from rivers have higher influence on susceptibility of coastal flooding. • 10% of the coastal areas of Bangladesh falls in high susceptible flood zones. [ABSTRACT FROM AUTHOR]