ANOVA Discriminant Analysis for Features Selected through Decision Tree Induction Method

Abstract

Feature selection plays an important role in recent years as the selected features improve the classification accuracy of the classifiers in software defect predictions. For improving the quality of the defect prediction model, effective features must be selected. Many feature selection algorithms have been developed for the improvement of defect prediction models. In this paper, we have used ANOVA Discriminant Analysis (ADA) to statistically prove that the features selected through Decision Tree Induction (DTI) approach are effective for defect predictions and only these features can be used for further use. ADA computes the means of variables and it selects the significant features from the groups iteratively for better results. By using ADA, it is found that the features selected through the DTI approach have higher discriminating power than others and the accuracy of the classifiers also increases when this feature set is used for many classifiers. Hence it is said that the selected feature set alone can be used for defect prediction instead of original feature set. It is also observed that the attributes selected through DTI are same as attributes used by ADA. Hence, these are proved to be significant. Wilk’s Lambda is taken as the significant measure. It shows the Discriminant power of the features i.e. discriminating power is high when Wilk’s lambda is low.