Optimized Feature Selection Approach Based on Entropy for Multi-Class Data Classification

Efficient machine learning heavily relies on solving the critical issue of feature selection, which serves as a valuable pre-processing method for improving data quality. Classification models face significant challenges, particularly when confronted with data containing irrelevant and redundant features. Unfortunately, only a few techniques specifically address feature selection in datasets with multi-class attributes. This research paper introduces a novel approach for feature selection called TSEFS, which utilizes entropy-based techniques such as fuzzy entropy and mutual information. TSEFS employs a two-stage process to select relevant features. The experimental results on three datasets demonstrate that the proposed method outperformed the existing feature selection methods.