<![CDATA[Stock price prediction is challenging due to fluctuating and nonlinear behavior. This study examines the effect of parameter optimization in Support Vector Machine (SVM) on prediction accuracy and error for stock prices. The dataset consists of PT Telekomunikasi Indonesia Tbk (TLKM) stock data from 2022–2024 obtained from Yahoo Finance. The workflow includes normalization, windowing-based feature construction, train–test splitting, and modeling using ε-Support Vector Regression (ε-SVR) with a Radial Basis Function (RBF) kernel. Parameter optimization is conducted via Optimize Parameters (Evolutionary) to find suitable C, gamma, and epsilon values, and the optimized model is compared against a baseline using LibSVM default parameters. Performance is evaluated using Root Mean Squared Error (RMSE), Absolute Error (AE), Correlation, and Prediction Average. Results indicate that the optimized model produces more stable predictions and follows the actual pattern more consistently, although the baseline may yield lower numerical error in some cases. This finding suggests that parameter optimization increases model sensitivity to training patterns but requires careful regularization to prevent accuracy degradation on test data.]]>
Copyrights © 2026
