Rainfall Forecast Analysis using Rough Set Attribute Reduction and Data Mining Methods
No 4/2014, December
Abstract
Developments in information technology has enabled accumulation of large databases and most of the environmental, agricultural and medical databases consist of large quantity of real time observatory datasets of high dimension space. The curse to these high dimensional datasets is the spatial and computational requirements, which leads to ever growing necessity of attribute reduction techniques. Attribute reduction is a process of reducing the data space by removing the irrelevant, redundant attributes from large databases. The proposed model estimates the enhancement achieved in spatial reduction and classifier accuracy using Rough Set Attribute Reduction Technique (RSART) and data mining methods. The first module of this proposed model has identified an efficient attribute reduction approach based on rough sets for spatial reduction. The next module of the proposed model has trained and tested the performance of Naive Bayes (NB), Bayesian Logistic Regression (BLR), Multi Layer Perceptron (MLP), Classification and Regression Tree (CART) and J48 classifiers and evaluated the accuracy in terms of each classification models. The experimental results revealed that, the combination of RSART based on Genetic Algorithm approach and Bayesian Logistics Regression Classifier can be used for weather forecast analysis.
Keywords
Attribute reduction, rough set, genetic algorithm, Bayesian Logistics Regression, rainfall forecasts, classification, accuracy.