Implementation of a Hybridized Machine Learning Framework for Flood Risk Management

Abstract

Annually, communities keep experiencing devastating effects and high fiscal loss due to flood risks resulting from climate change, severe rainfall, rapid population growth, urbanization and poor governance. The management of flood risks is limited due to inadequate information and awareness of hazard. Hence, the import of exploring realistic flood risk mitigation measure become very paramount. This paper implements a hybridized flood risk analytic framework, consisting of supervised and unsupervised machine learning methodologies, for the discovery of useful knowledge, clustering and prediction of flood risks of severity level. MatLab and Tanagra software characterized the programming environment while data obtained from Nigeria Emergency Management Agency was used to evaluate the system. K-means based on silhouette criterion discovered 2, 3 and 4 as top performing numbers of clusters out of 19 clusters. Based on human experts’ judgment four clusters was chosen with Squared Euclidean distance the best representative of the four distance measures. The Self Organising Map (SOM) provided a visual of the input attribute within the four different groups with similar patterns. The supervised learning procedure via Adaptive Neuro Fuzzy Inference System (ANFIS) to predict based on resultant dataset with for flood severity level was performed. ANFIS for flood risk classification optimized by Genetic Algorithm produced Root mean squared error of 0.3225 and Error Standard Deviation of 0.4080. The perception of emergency risk management is very important; therefore this research demonstrates its practical application, data mining techniques and tools for emergency risk management.