Machine learning approach for flood risks prediction

Abstract

Flood is one of main natural disaster that happens all around the globe caused law of nature. It has caused vast destruction of huge amount of properties, livestock and even loss of life. Therefore, the needs to develop an accurate and efficient flood risk prediction as an early warning system is highly essential. This study aims to develop a predictive modelling follow Cross-Industry Standard Process for Data Mining (CRISP-DM) methodology by using Bayesian network (BN) and other Machine Learning (ML) techniques such as Decision Tree (DT), k-Nearest Neighbours (kNN) and Support Vector Machine (SVM) for flood risks prediction in Kuala Krai, Kelantan, Malaysia. The data is sourced from 5-year period between 2012 until 2016 consisting 1,827 observations. The performance of each models were compared in terms of accuracy, precision, recall and f-measure. The results showed that DT with SMOTE method performed the best compared to others by achieving 99.92% accuracy. Also, SMOTE method is found highly effective in dealing with imbalance dataset. Thus, it is hoped that the finding of this research may assist the non-government or government organization to take preventive action on flood phenomenon that commonly occurs in Malaysia due to the wet climate.