Predicting stock prices using linear and non linear machine learning models

Abstract

Predicting stock price movements is inherently difficult due to market volatility and the influence of numerous external factors. This study develops a machine learning framework that leverages historical opening prices to forecast short-term stock prices for selected publicly traded companies. Using five years of daily data, the model incorporated features from three consecutive opening prices to predict the subsequent day’s opening price. Four machine learning models were trained and evaluated: Linear Regression, Decision Tree, Random Forest, and Neural Network. Performance was assessed using mean squared error (MSE), with the Random Forest model achieving the lowest error, followed closely by the Neural Network. An ensemble approach that combined model predictions yielded a slight further reduction in error. To illustrate potential applications, a simple trading simulation was conducted using linear regression predictions, which showed that under ideal conditions a $500 investment in Microsoft stock could grow substantially. While the models demonstrated only modest predictive accuracy, the limited feature set constrains their ability to generalize. Future research should investigate richer input features, advanced validation techniques, and hyperparameter optimization to improve forecasting reliability.