Machine learning techniques for flood forecasting

ABSTRACT Climate change resulted in dramatic change in the monsoon precipitation rates in Malaysia, contributing to repetitive flooding events. This research examines different substantial practicalities of machine learning (ML) in performing high-performance and accurate FF. The case study was The Dungun River. IGISMAPs datasets of water level and rainfall were investigated (1986–2000). The Forecasting was implemented for current (1986–2000) and near future (2020–2030). ML algorithms were Logistic Regression, K-Nearest neighbors, Support Vector Classifier, Naive Bayes, Decision tree, Random Forest, and Artificial Neural Network. Simulations were run in the Colab software tool. The results revealed that between 1986 and 2000, there would be an average of (18–55) floods around the Dungun River Basin. Floods occurred rarely before 1985. They have been common since 2000. 35 floods occurred annually on average since 2000. It is predicted that between 2020 and 2030, flooding events would grow on the Dungun River Basin. Most floods occurred due to rainfall between 1 and 500 mm. The maximum frequency of flooding was measured at 110 occurrences at a rainfall of 250 mm. The overall accuracies were 75.61%/ random forest, 73.17%/ KNN, and logistic regression/ 48.78%. Overall, the ANN models had a competitive mean accuracy of 90.85%.