Optimization of Feature Selection using Firework Algorithm for Machine Learning Algorithm

Abstract

In machine learning and deep learning, the optimal feature selection plays an important role in enhancing performance, decreasing computational cost, and improving the interpretability of the algorithm. The performance of the machine learning algorithm is impacted by the noisy, redundant, and meaningless features found in the majority of classification problem datasets. Feature selection is the process of selecting a feature subset and a search method for finding the optimal subset of features from many features using a fitness function to improve the accuracy and execution time. This research focuses on the application of the fireworks algorithm as a useful tool for optimizing feature selection. By navigating the feature space, the recommended technique finds the optimum subset of features that maximize a model's performance. By analyzing the fitness function, which combines the complexity of the model with the predictive ability of the chosen features, the program repeatedly improves the feature subset. The paper makes use of widely known datasets on breast cancer that include a limited number of characteristics. The classification performance of selected feature subsets is evaluated using classification techniques: Support vector machine, logistic regression, and bagging classifier. The proposed algorithms are better than the particle swarm optimization algorithm, ant colony optimization algorithm, principal component analysis, and so on. The results indicate that a certain feature subset may be chosen with higher accuracy through the use of the recommended techniques, as opposed to using all characteristics. An optimal subset of feature selection techniques enhances the accuracy and decreases the number of features. Enhancements in classification accuracy are supported by a substantial decrease in the number of features with a higher weight on the fast reduction of the fitness function.