Dimensionality Reduction by Cascading Mutual Correlation with Symbolic Approach

In this paper, we propose a novel cascading approach, by cascading the feature selection method using mutual correlation with this symbolic approach. In the symbolic approach, the new dimensionality reduction method through transformation of features into symbolic data using the property of collinearity and variance based cumulative sum of features is used. The feature values are transformed into line segments and thus reduced to two symbolic features namely, number of line segments and average slope of the line segments. In addition the first and last feature values are also considered to distinguish the samples with the same average slope values. In this proposed approach of cascading the feature selection method using mutual correlation with this symbolic approach, the entire feature set is reduced to only 4 features. Experimental results on the standard datasets WDBC, WBC, CORN SOYANEAN and WINE shows that the proposed methods achieve better classification performance with negligible time.