The Effect of Mutual Information-Based Feature-Selection on Support Vector Regression for Goto Stock Price Prediction
Abstract
The dynamic movement of stock prices makes price prediction an important aspect in supporting investment decision-making. This study aims to apply Support Vector Regression (SVR) with Mutual Information (MI)-based feature selection to predict the stock price of PT. GoTo Gojek Tokopedia Tbk (GOTO). The data used are historical GOTO.JK data from 2 January 2023 to 30 December 2025, obtained from Yahoo Finance. After preprocessing, 688 observations were obtained and divided chronologically into 550 training data and 138 testing data. Next, 32 features were constructed, consisting of historical prices, trend indicators, momentum, volatility, volume, as well as lag and return features. Feature selection was performed using MI to measure the degree of association of each feature with the next period's closing price, and several scenarios were then formed based on the MI ranking. Modeling was carried out using SVR, while evaluation used RMSE, MAE, MAPE, and Directional Accuracy (DA). The results show that the Top 5 MI scenario, consisting of Close, Low, High, the 5-period EMA, and Open, yields an RMSE of 1.5774, an MAE of 1.1203, a MAPE of 1.8342%, and a DA of 48.5507%. These results indicate that MI-based feature selection helps SVR exploit more informative features and achieve effective predictive performance with fewer features. Accordingly, Top 5 MI was selected as the best feature configuration for SVR modeling in GOTO stock price prediction.
Keywords
Share and Cite
Article Metrics

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.

