Evolutionary feature selection

Feature selection is an approach to selecting the best set of features from a feature pool. Its goal is to increase the performance of the machine learning model by providing sufficient information while avoiding redundant or irrelevant features. Due to the high dimensionality of data in practical problems, solutions ranging from genetic algorithms to reinforcement learning have been recently tried to solve this task. In this work, we propose a novel feature selection architecture that uses metaheuristic techniques combined with evolutionary algorithms and chaos theory to select the best features for a model. It uses the concept of evolution, which guides the algorithm to the best path and a chaotic map function to create new random subsets of features. The backbone of this algorithm, mutation and crossover operator, is inspired by genetic algorithms. It uses these methods to increase the exploration and exploitation strategies for the search space. We tested the proposed method on 10 datasets using different machine learning models and achieved significant improvement on each dataset compared to other methods in the literature.