Spatial prediction of flood susceptible areas using machine learning approach: a focus on west african region

Abstract

The constant change in the environment due to increasing urbanization and climate change has led to recurrent flood occurrences with a devastating impact on lives and properties. Therefore, it is essential to identify the factors that drive flood occurrences, and flood locations prone to flooding which can be achieved through the performance of Flood Susceptibility Modelling (FSM) utilizing stand-alone and hybrid machine learning models to attain accurate and sustainable results which can instigate mitigation measures and flood risk control. In this research, novel hybridizations of Index of Entropy (IOE) with Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF) was performed and equally as stand-alone models in Flood Susceptibility Modelling (FSM) and results of each model compared. First, feature selection and multi-collinearity analysis were performed to identify the predictive ability and the inter-relationship among the factors. Subsequently, IOE was performed as bivariate and multivariate statistical analysis to assess the correlation among the flood influencing factor’s classes with flooding and the overall influence (weight) of each factor on flooding. Subsequently, the weight generated was used in training the machine learning models. The performance of the proposed models was assessed using the popular Area Under Curve (AUC) and statistical metrics. Percentagewise, results attained reveals that DT-IOE hybrid model had the highest prediction accuracy of 87.1% while the DT had the lowest prediction performance of 77.0%. Among the other models, the result attained highlight that the proposed hybrid of machine learning and statistical models had a higher performance than the stand-alone models which reflect the detailed assessment performed by the hybrid models. The final susceptibility maps derived revealed that about 21% of the study area are highly prone to flooding and it is revealed that human-induced factors do have a huge influence on flooding in the region.