This study aims to propose a genetic algorithm (GA)-based feature reduction (FR) approach applicable to any classification data set, regardless of domain, dimensionality and size of the data set, with any number of classes and data format.
Two levels of experiments were conducted to evaluate the efficacy of the proposed model. Level-I focuses on assessing the reduction percentage of features, classification accuracy (CA) and learning time (LT) using the J48 decision tree learner. Level-II aims to examine the improvement in performance metrics – specifically CA, LT and the maximum achieved CA – across 18 data sets, using four diverse and competitive classifiers, namely, naïve Bayes, artificial neural network, K-nearest neighbour and random forest, both with and without FR.
The proposed GA-based model achieves a significant reduction in the number of features across the data sets. Furthermore, it leads to a noticeable decrease in LT and an improvement in CA across various classifiers on the refined data sets.
Experimental results demonstrate that the proposed model is scalable and performs at par with other specialized FR techniques.
