Comparing performance of machine learning algorithms in a flood prediction model with real data sets

Abstract

Flood is one of an unforeseen and often sudden event or a situation that causes severe damage, destruction and human suffering which requires help by requesting to national or international level. The implementation of machine learning approaches in flood prediction may reduce all the risk factors. Machine learning is one of the method that provide better performances and it is cost-effective and recently used among hydrologists. However, the capability of each machine learning algorithm is different for each type of tasks which is called generalization problem. Thus, for this research, three machine learning methods which are Artificial Neural Network (ANN), Support Vector Machine (SVM) and Decision Tree (DT) is chosen for flood prediction model. Each of machine learning algorithms are built and trained in order for they to work accordingly with two different datasets. The aim of this research project is to investigate the performance of three selected machine learning algorithms and compared their accuracy. ANN has shown promising results with the highest performance accuracy of 98% in dataset 1 and 77.10% in dataset 2.