Statistical Feature Ranking and Fuzzy Supervised Learning Approach in Modeling Regional Rainfall Prediction Systems

Rainfall prediction is an essential and challenging task in hydro-meteorology. Most of the existing weather dataset used for prediction consists of observatory record of several atmospheric parameters. Identifying the significant parameters from irrelevant and redundant parameter set for weather prediction is important because irrelevant parameters may decrease the prediction accuracy. The main intent of this research is to identify the influencing weather parameters for improving daily rainfall forecast efficiency. A parameter selection module identifies the significant parameter based on information gain based feature ranking. Fuzzy supervised learning module evaluates the performance of fuzzy classifiers before and after parameter selection. In the evaluation phase, learning techniques was analyzed in terms of Accuracy Rate (AcR), Root Mean Squared Error (RMSE) and Misclassification Rate (McR). Experimental results revealed that, parameter subset selection has significantly improved the performance of the learning techniques. The investigation results identified minimum temperature, relative humidity and evapotranspiration as influencing weather parameters for rainfall prediction. Empirical results revealed Fuzzy Unordered Rule Induction Algorithm (FURIA) as a suitable rainfall prediction approach. This fuzzy model achieved an enhanced accuracy rate of 84.10% after parameter selection with nominal misclassification rate of 0.1590%.