Urban Pluvial Flood Prediction Using Machine Learning Models

Urban pluvial flooding causes threats to human lives, economy, infrastructure, and ecosystems around the world, particularly in developing countries. Pakistan is among those developing countries that have become vulnerable to urban flooding due to global warming, climate change risks, rapid urbanization, and the occurrence of extreme and unusual rainfall events in the monsoon season. To effectively mitigate the adverse effects of pluvial flooding, there is a pressing need for an advanced and accurate predictive model that is capable of predicting urban pluvial flooding. The aim of this research is to utilize the potential of machine learning (ML) techniques and algorithms to accurately predict the occurrence of urban pluvial flooding, for this study we have worked on the monsoon rainfall data of multiple cities of Sindh province in Pakistan. We made an attempt to use multiple ML models especially logistic regression (LR), random forest (RF), k nearest neighbor (KNN), and support vector machine (SVM) to propose the best and accurate predictive model. The developed models were evaluated and assessed using various performance evaluation metrics which includes accuracy, recall, precision and F1 score. The best among proposed models was created using LR which achieved an accuracy of 97%, while the accuracy of other models which were created using SVM, KNN and RF were 96%, 93%, and 92% respectively.