Improved Support Vector Machine (SVM) Performance on Go-Jek Service Review Classification Using Particle Swarm Optimization (PSO)
Windha Mega P.D., Haryoko · 2022 International Conference on Informatics, Multimedia, Cyber and Information System (ICIMCIS) · 2022
The choice of parameters in the classification process using the Support Vector Machine significantly affects the optimization of the results. Continuing research that has been done before, by optimizing SVM parameters using genetic algorithms that produce accuracy up to 86%. From the experimental results, it has been proven that genetic algorithms can be used to optimize parameters in SVM, but the number of iterations in GA is quite a lot and requires a longer computational time to find a solution. There is a more popular optimization method, PSO. Where PSO has a more efficient calculation than the Genetic Algorithm. For this reason, this research will apply PSO as an optimization method for SVM parameters as a classifier. PSO-SVM performance can improve the previous accuracy using GA-SVM. The results of the test are carried out with accuracy, and the F1 score test to determine the results of the application of the PSO Algorithm in SVM. Furthermore, it was compared to the accuracy of the kernel, which is the best. At this stage, it can be concluded that the accuracy of PSO-SVM increased to 87% in the Poly kernel with faster computing time.