SVR PARAMETERS OPTIMIZATION USING PSO AND WAPSO IN STOCK PRICE PREDICTIONS

Abstract

This study discusses the optimization of Support Vector Regression (SVR) parameters to predict stock prices using Particle Swarm Optimization (PSO) algorithm and Weight Attribute Particle Swarm Optimization (WAPSO) algorithm to improve SVR accuracy and to determine the effect of the inertia weight attribute in the optimization. WAPSO is a Particle Swarm Optimization algorithm with the addition of an inertia weight attribute. The implementation of the SVR, SVR-PSO and SVR-WAPSO algorithms is carried out on three LQ45 stock data, namely TLKM, BBRI and ADRO. The results of the accuracy of the implementation of the SVR algorithm are 79.02493% (TLKM), 67.83047% (BBRI), 88.94952% (ADRO), the SVR-PSO algorithm is 98.64916% (TLKM), 98.32181% (BBRI), 97.90267% (ADRO) and the SVR-WAPSO algorithm is 98.64921% (TLKM), 98.32496% (BBRI), 97.89889% (ADRO). Based on statistical tests in 3 data, it is concluded that the weight attribute has an effect in increasing the mean fitness value so that WAPSO has a better mean fitness value. The difference in the increase in the mean fitness value at WAPSO and PSO on TLKM data is 0.1230, BBRI data is 0.2202, and ADRO 0.0597.