Feature Selection Method Based on Classification Performance Score and P-value

Abstract

Learning from high dimensionality data has gained increasing attention in recent years due to the massive growth of skewed data across many scientific fields. Some researches show that feature selection does play an essential role in selecting the significant features, also reducing the number of dimensions in the process.

In this study, we propose a feature selection approach based on P-value and a performance score. In fact, our solution contains two main steps that work in parallel. The first one measured the performance score of each feature applying support vector machine (SVM). The second estimated the P-value of each feature, then fixed the threshold. Combining both steps, we obtain the feature kth and the range features that will be used to create the subsets. We tested the classification performances of the different selected features subsets using performance score with three different classifiers, and we choose NB that showed the highest performance to continue the discussion section based on it. Besides, we assessed the performance score of the proposed approach with four different datasets from Kaggle. According to the results of the study, and comparing two cases for each selected subset, we concluded that the higher performance could be reached by including the Kth, Kth + 1 in the selected subset of features.