Machine learning application to predict flood zones in Togo

Climate change has led to an increase in the frequency of extreme and sudden precipitation events in cities, resulting in significant losses from natural disasters. Every year, citizens in West African states, specifically Togo, face heavy rains that often lead to flooding. To mitigate the negative impacts, it is essential to identify the factors determining the occurrence of floods and the most flood-prone areas. But how can we determine if a point or an area is susceptible to flooding? One possible approach is to perform Flood Susceptibility Modeling (FSM), which utilizes machine learning models to obtain accurate and sustainable results. In this study, we propose the following models: Artificial Neural Networks (ANN), Logistic Regression (LR), Support Vector Machine (SVM), and Random Forest (RF) to compare their performances as standalone models in flood susceptibility modeling. Prior to constructing the models, we conducted feature selection and multicollinearity analysis to identify the most predictive factors and their inter relationships. The results showed that the RF model produced higher performance and prediction rates compared to the LR model, achieving 94.74% and 69.64% accuracy, respectively. In percentage, the other models achieved performances of 77.27% for SVM and 82.11% for ANN, respectively. Furthermore, the models highlighted that the most flood-prone areas are low-lying regions, devoid of vegetation, and located near roads. They also revealed that anthropogenic factors have a significant impact on flooding in the study area. By utilizing these machine learning techniques to predict flood prone areas, our study provides valuable insights for formulating flood prevention policies and plans.