Comparative Analysis of Traditional and Optimization Algorithms for Feature Selection

Machine learning enables the automation of the system to generate results without direct assistance from the environment once the machine is trained for all possible scenarios. This is achieved by a series of processes such as collecting relevant data in raw format, exploratory data analysis, selection and implementation of required models, evaluation of those models, and so forth. The initial stage of the entire pipeline involves the necessary task of feature selection. The feature selection process includes extracting more informative features from the pool of input attributes to enhance the predictions made by machine learning models. The proposed approach focuses on the traditional feature selection algorithms and bio-inspired modified Ant Colony Optimization (ACO) algorithm to remove redundant and irrelevant features. In addition, the proposed methodology provides a comparative analysis of their performances. The results show that the modified ACO computed fewer error percentages in the Linear Regression Model of the dataset. In contrast, the traditional methods used outperformed the modified ACO in the SVR model.