A novel feature selection framework for improving detection performance of supervised classifiers

This research aims to develop a novel feature selector for improving the detection performance of supervised classifiers. Handling large number of features is a tedious process. One solution is to select only the relevant features and eliminate both irrelevant, redundant features from the original set. A new feature selection method based on Class Conditional Probability (CCP) is proposed in this research. The CCP for every attribute is calculated using Naive Bayes approach. The related attributes which has the CCP value greater than the threshold value is selected as relevant features. Then, the reduced feature set is applied to different classifiers such as C4.5, Naive Bayes (NB), Support Vector Machine (SVM), Nearest Neighbour (NN) and K-Nearest Neighbour (K-NN). Different datasets from UCI repository are considered to prove the efficacy of the proposed feature selector based on the number of selected features, time taken to build the model and classification accuracy.