A Bayesian Approach to Prediction of Flood Risks

Abstract

Flood is a temporary overflow of a dry area due to overflow of excess water, runoff surface waters or undermining of shoreline. In 2014, Malaysia grieved with the catastrophic flood event in Kuala Krai, Kelantan, which sacrificed human lives, public assets and a total of RM 2 billion loss. Due to uncertainties in flooding event, this research is set to compare three variations of Bayesian approaches in classifying the risk of flood into two classes; flood or no flood. The study involved data from Kuala Krai, which serves as the main observation point. The dataset contains six attributes, which are water level, rainfall daily, rainfall monthly, wind, humidity, and temperature. The classification experiment will be conducted using three variants of Bayesian approaches, which are Bayesian Networks (BN), Naïve Bayes (NB), and Tree Augmented Naive Bayes (TAN). The outcomes of this research will show the best algorithm performance in term of accuracy for three Bayesian-based learning prediction algorithms. In the future, this prediction system is hoped to assist related agencies in Malaysia to categorize land areas that face high risk of flood so preventive actions can be planned in place.   Â