Positive and Negative Feature-Feature Correlation Measure: AddGain

Abstract

Feature selection techniques are searching for an optimal subset of features required in the machine learning algorithms. Techniques like the statistical models have been applied for measuring the correlation degree for each feature separately. However, the mutual correlation and effect between features is not taken into consideration. The proposed technique measures the constructive and the destructive effect (gain) of adding a feature to a subset of features. This technique studies feature-feature correlation in addition to the feature-class label correlation. The optimality in the resulted subset of features is based on searching for a highly constructive subset of features with respect to the target class label. The proposed feature selection technique is tested by measuring the classification accuracy results of a data set containing subsets of constructively correlated features. A comparative analysis shows that the resulted classification accuracy and number of the selected feature of the proposed technique is better than the other feature selection techniques.