Adjusting parameters of one-sided control chart-based support vector machines using genetic algorithm and particle swarm optimization

Tóm tắt

This study examines parameter optimization for one-sided control chart-based Support Vector Machines (SVM) using Genetic Algorithm (GA) and Particle Swarm Optimization (PSO). Statistical learning classification effectively addresses statistical process control issues, especially in managing big data complexities. While SVM has computational benefits, its predictive accuracy relies on proper parameter selection. This research presents two methods—D-SVM-GA and D-SVM-PSO—to optimize key parameters like moving windows ( ), the Gamma (σ²), and the penalty cost (C). Results show upper control limits of 0.518 for D-SVM-GA and 0.515 for D-SVM-PSO, indicating effective optimization. The  values highlight improved detection performance, where higher  results in lower , enhancing responsiveness to shifts. These findings underscore the effectiveness of evolutionary algorithms in optimizing SVM parameters, offering insights for future applications in process control and statistical learning, with potential for improved accuracy and efficiency in data-driven decision-making.