| Abstrakt: |
In machine learning models, feature selection plays a crucial role. It reduces overall data, minimizes storage requirements, and enhances algorithm performance. Despite this, greedy and exhaustive search methods may not be optimal as the number of features increases. Metaheuristic algorithms are a more sensible way to deal with this issue. In this study, the Grasshopper Optimization Algorithm (GOA) is applied to feature selection. Although GOA may be relatively easy to implement, it may not fully leverage each iteration and may become stuck in local optima. The comfort zone in GOA influences the grasshopper movement within the search space, influencing exploration and exploitation. As a constant, the algorithm changes the comfort zone linearly. The proposed algorithm, Signature Chaos GOA (SCGOA), overcomes these limitations in several ways. Firstly, it constructs the initial population using correlations. Second, unlike existing methods, it specifies specific procedures for initial and final iterations. After the initial iteration, the algorithm adjusts the comfort zone parameters dynamically using chaos theory and fuzzy signatures. Lastly, SCGOA aims to optimize both Support Vector Machine (SVM) parameters and feature subsets simultaneously. Objective functions include classification error, the proportion of selected features, and redundancy. In addition, different algorithms such as the Firefly Algorithm (FA), the Bat Algorithm (BA), and the Particle Swarm Optimization (PSO) are compared. In comparison with FA, BA, PSO, and GOA, the proposed algorithm can improve the objective function by 30.6%, 34.9%, 7.6%, and 33.3%, respectively. [ABSTRACT FROM AUTHOR] |
Full text from SpringerLink