DENSITY-BASED FEATURE EXTRACTION TO IMPROVE THE CLASSIFICATION PERFORMANCE IN THE DATASETS HAVING LOW CORRELATION BETWEEN ATTRIBUTES

Abstract

If there is low correlation between attributes belonging to the same classes of datasets, the success rates of classification methods will be low. The aim of this study is to increase the success rates of classifiers in such datasets. In this study, attributes of datasets were firstly converted to density coefficients. Thus, the new datasets having higher correlation between the attributes have been created. Then compared to the structure of the original dataset, this new dataset was evaluated in terms of contribution to classification performance. For the evaluation process, various classification methods were applied to original datasets as well as new datasets generated by the proposed method. According to comparison results, it was observed that the proposed method contributes to the classifier performance about 17%.