Assessing deluge predictability and deterministic attributes of artificial learning systems

Natural disasters including flood cannot be accurately predicted both temporally and geographically. Further the extent to which they would disrupt us socially is usually unforeseeable, though great economic devastation has often been entailed. Much effort has thus been spent on addressing the issue, most notably is on implementing flood prediction models based on off-the-shelf artificial learning paradigms. Despite sound academic values, the applicability of these models remains hypothetical, mainly due to limited in actu scrutiny. This paper revisits those techniques and studies their predictability and extracted deterministic attributes. Our main contribution is benchmarking them with recent deluge, in 2011, with the full scale census and comprehensive GIS survey. Resultant ramifications are not only relative ranking amongst the opted candidates, i.e., ANN, GA, Fuzzy Logic and SOM, based on mere forecasting accuracy and generalization ability, but also the rationale behind predictive attributes, which in turn serves as the guidelines and precautions on applying these prominent tools on wider range of the geographical scenarios.