Une méthode hybride basée sur l’information mutuelle et les algorithmes génétiques pour la sélection des attributs

Abstract

Feature selection is a crucial step in the machine learning process, aiming to identify and select the most informative features from an original dataset. Among the techniques used in this process, we find MI, CH2, IGI, etc. Despite their effectiveness, these techniques suffer from the drawback of redundancy, resulting in poor classification model performance. In this work, we opted for a hybrid approach based on the IG method and genetic algorithm. Firstly, we use IG to evaluate the relationship between each feature and the class variable. Features with high IG scores are considered more discriminative. Then, we employ a genetic algorithm to search within the space of features selected by IG and find an optimal subset using operations such as selection, crossover, and mutation. The experimental results confirm that our hybrid method has achieved our objective by improving performance, significantly reducing redundancy, and outperforming other methods.