Comparative Analysis and Ensemble Optimization of Machine Learning Models for Next-Day Stock Price Prediction

Predicting stock market movements remains a complex challenge due to the market’s inherent volatility, noise, and sensitivity to both fundamental information and investor behavior. This study investigates the application of multiple machine learning models for next-day stock price prediction by utilizing five years of historical daily opening price data obtained from Yahoo Finance. A rolling three-day input window was used to forecast the fourth day’s price. Five models– linear regression, decision tree regression, random forest regression, multi-layer perceptron (MLP), and k-nearest neighbors (KNN) –were trained and tested using a training-testing split, with early stopping implemented to reduce overfitting and improve generalization. Model performance was assessed based on mean squared error (MSE). Ultimately, the simulation demonstrated consistent profitability.
The results show that while particular models exhibit varying levels of accuracy, combining models can improve robustness and reduce bias towards certain companies. However, there are still limitations, particularly due to the reliance on historical price data and the inability to fully capture external shocks, current events, and structural changes in financial markets. These findings highlight both the potential and the constraints of pure quantitative machine learning in the stock prediction field, suggesting that future approaches should incorporate more adaptive and qualitative aware methodologies.