Effective rainfall prediction using hybrid intelligent systems

Abstract

Rainfall prediction is a multidimensional, dynamic, data-intensive and chaotic process. It is one of the several computational tasks identified by meteorologists as the core problem in weather modeling across the globe. This investigation aims to determine the significant atmospheric parameters and appropriate predictive techniques to enhance the precision of weather prediction system using hybrid data-driven approach. The remarkable contribution of this thesis is to improve the efficiency of the data-driven approach by exploiting the empirical characteristics of the input weather parameters. The research outcomes reveal that, the fundamental and generic data mining techniques are ineffective to comprehend the hidden input-output relationship. Therefore, this research introduces an optimal feature selection using Maximum Frequency Weighted Reduct Selection (MFWRS) and Reduct Selection Using Genetic Algorithm (RSGA) to identify the effective input parameters in modeling short-term rainfall forecast scenario. The thesis further outlines the application of different intelligent computing approaches based on rough sets, fuzzy sets, evolutionary computing, neural networks and their implications for the practical workings of short-term rainfall forecasting. The proposed hybrid frameworks modeled using Adaptive Neuro-Fuzzy Inference Systems, Fuzzy Rule-Based Classification and contemporary data mining techniques performed substantially better when trained with feature subsets generated using the proposed optimal reduct selection techniques. The thesis introduces an unconventional Adaptive Rough-Evolutionary Neuro Approach (ARENA) and Adaptive Rough Neuro-Fuzzy Approach (ARNFA) based hybrid intelligent systems. The proposed ARENA achieved 98.01% prediction accuracy with a nominal error rate of 1.99%. The methodical analysis revealed ARENA as an appropriate tool to deal with the real-world rainfall prediction efficiently by means of identifying significant weather parameters.