Improved Discriminatory Ability using Hybrid Feature Selection via Approach Inspired by Grey Wolf Search and Ensemble Classifier for Medical Datasets

Abstract- Medical datasets inevitably suffer from redundant and irrelevant attributes, which reduce data mining algorithms' ability and often lead to uninterpretable results. Therefore, the first step in medical diagnosis problems is to reduce dimensionality. This paper presents a computational method that takes advantage of wrapper subset evaluation with a meta-heuristic algorithm in a two-phase process to improve the classification performance with a group of meta-classifiers. The first phase filters the feature domain using the information gain ratio in an attribute evaluation method. The first layer's output serves as an input feature for the second phase, which uses grey wolf optimization to find the optimal feature space. An ensemble-based classifier scheme was built based on C4.5, RandF, and ForestPA classifiers to obtain the final decision. The proposed method was validated on several medical datasets of the UCI Machine Learning Repository. The results show that the suggested approach yields significant performance improvements in Accuracy performance compared to other classifiers. Keywords- Data Mining; Feature Selection; Grey Wolf Optimization; Information Gain Ratio; Ensemble Classifier