Abstract

Feature selection is a method in Data mining to reduce the features from the original dataset by removing the noisy features from the dataset to improve the performance of the classifiers in terms of improving the prediction accuracy. Classification tasks often include, among the large number of features to be processed in the datasets, many irrelevant and redundant ones, which can even decrease the efficiency of classifiers. Feature Selection (FS) is the most common pre-processing technique utilized to overcome the drawbacks of the high dimensionality of datasets. The proposed SEF-USIEF Feature Selector: An approach to Select Effective Features and Unselect Ineffective Feature methods increases the performance of the classification methods by eliminating the irrelevant features from the dataset. This proposed SEF-USIEF method is implemented for the numerical datasets. This method is derived from the ward Minimum Variance cluster method and in this experiment Minimum Variance is used as the feature selector method. The numerical datasets are obtained from the UCI repository and WebKB repository. The results obtained by the proposed SEF-USIEF method are compared with the existing feature selection methods to analyze whether the features are reduced by the SEF-USIEF method or not. Then, features are given as input to the classifiers to check whether the classifier performance is increased or not. Based on the compared analysis of the results, the SEF-USIEF method proved that the performance of the classifiers is increased, and also the selected features are reduced when compared to the existing feature selection methods.