Enhancing Stock Prediction Accuracy Using Hybrid Lasso-ARBF-SVR with Adaptive Kernel for Indonesian Banking Stocks

Abstract—Predicting stock prices in emerging markets like Indonesia presents significant challenges due to high volatility, data noise, and complex macroeconomic interdependencies. This paper introduces a novel hybrid model, Lasso-ARBF-SVR, that combines Lasso regression for sparse feature selection with an Adaptive Radial Basis Function (ARBF) kernel-based Support Vector Regression (SVR) for nonlinear modeling. The ARBF kernel dynamically adjusts its bandwidth based on local data density, enabling better adaptation to market regime changes. We evaluate our model using daily stock data from five major Indonesian banks (BBRI, BBNI, BMRI, BBCA, BBTN) from 2010 to 2024. Experimental results demonstrate that the proposed model achieves unprecedented accuracy with a Mean Absolute Percentage Error (MAPE) of 0.78% ± 0.05%, R2 of 0.997 ± 0.001, and Root Mean Square Error (RMSE) of 45.2 ± 2.1, significantly outperforming state-of-the-art benchmarks including XGBoost, Random Forest, LSTM, and conventional SVR (p < 0.001 via Diebold-Mariano test). An ablation study confirms the contribution of each component: Lasso feature selection reduces MAPE by 31%, while the ARBF kernel provides an additional 22% improvement. The model maintains robust performance during market crises, including the COVID-19 pandemic, and achieves real-time prediction capability on edge devices with a throughput of 38 predictions per second on Raspberry Pi 5. Index Terms—Banking stock prediction, hybrid Lasso-ARBFSVR, adaptive kernel, sparse feature selection, financial machine learning, emerging markets