Stock Price Prediction: A Machine Learning Approach

Abstract

Accurately predicting stock prices remains a complex task due to market volatility influenced by economic indicators and geopolitical events. This study presents a Hybrid Stock Sequence Learner (HSSL) model that integrates Support Vector Machine (SVM), Support Vector Regression (SVR), and Linear Regression (LR) to improve forecasting performance in dynamic financial environments. The model employs an attention-gated mechanism and regularization to capture higher-order feature interactions while preventing overfitting. SVM captures non-linear patterns, LR enhances interpretability by calculating feature impacts, and SVR enables adaptive modelling through high-dimensional feature mapping. The HSSL model was empirically evaluated using historical stock data from five entities, namely Apple, Microsoft, Walt Disney, Alphabet, and the S&P 500 index. This data was sourced from Yahoo Finance. Results show that HSSL achieves a Mean Squared Error (MSE) of 4.414 and a Root Mean Squared Error (RMSE) of 29.843, outperforming baseline models. These results demonstrate that our model effectively reduces prediction errors and captures market trends. Our proposed approach offers a robust and interpretable forecasting framework suitable for short-term stock price prediction and decision-making in highly volatile markets, serving as a vital tool for stock investors to evaluate potential risks and adjust strategies to minimize losses.