Improved Quantum Genetic Algorithm with Mutation – Study on Algorithm Convergence
Agnieszka Wielgus, Dawid Podgórski, Urszula Libal
This paper investigates the impact of key parameters on the convergence and performance of an Improved Quantum Genetic Algorithm (IQGA) applied to Support Vector Machine (SVM) kernel optimization. SVMs are powerful classification tools whose performance is heavily reliant on appropriate parameter selection, a process often inefficient with traditional optimization techniques. Quantum Genetic Algorithms (QGAs) offer a promising alternative, exhibiting improved convergence and accuracy compared to classical Genetic Algorithms, though they can still be susceptible to local optima. Building upon previous work with IQGA, this study focuses on a detailed analysis of parameter influence and introduces two novel mutation schemes to further enhance population diversity. Specifically, the research examines the effects of population size, number of subpopulations, crossover iteration interval, and rotation range on the convergence behavior and final accuracy of the IQGA-SVM. Numerical experiments were conducted on the Wine and Breast Cancer Wisconsin benchmark datasets. The results demonstrate that while population size significantly affects convergence speed, it does not directly translate to improved final accuracy. Furthermore, the study reveals the positive influence of a larger number of subpopulations and less frequent crossover iterations on convergence. The proposed M2 mutation strategy yielded the best overall performance. Importantly, the IQGASVM consistently achieved high mean test accuracy across various parameter settings, significantly outperforming a standard SVM with RBF kernels trained under identical conditions. These findings highlight the potential of IQGA-SVM for efficient kernel optimization, particularly its ability to achieve high accuracy with smaller population sizes, a crucial factor for practical quantum computing implementations.
IEEE EUROCON - International Conference on Computer as a Tool (EUROCON)•Jun 4, 2025•1 citation