Stock Price Forecasting with Support Vector Regression Using Firefly Algorithm Optimization

Abstract

Stock prices have a fluctuating nature; that is, they can change in a short time. Therefore, it is necessary to analyze and anticipate the risks that will occur by forecasting stock prices. The method that can be used to predict stock prices is Support Vector Regression (SVR), which has the advantage of not requiring certain assumptions to be used, being able to overcome overfitting, training time is faster, and being able to predict time series-based data such as stock prices. However, because the parameters are difficult to determine, SVR requires the help of an optimization method to find the optimal parameters, namely the Firefly Algorithm (FA) method. The combination of SVR-FA is also considered to have the advantage of producing a smaller error value than the combination of other methods. The stock data used is PT's daily stock data. Indofood Sukses Makmur Tbk. and USD-IDR exchange rate data from January 1st, 2012, to January 31st, 2022. This study aims to obtain information on the accuracy of results in forecasting the stock price of PT through the best combination of values ​​and several parameters. The best accuracy results are obtained by combining 100 SVR iterations, 10 FA iterations, and 40 individual firefly numbers with a MAPE testing accuracy of <1% and 0.6796%, which can provide good forecasting results.