Comparative Analysis of ML models for Stock Price Prediction

Due to the unstable and non-stationary nature of financial market volatility, accurately predicting stock prices remains a challenging task. This study employs historical prices and technical indicators as features, comparing five machine learning models: linear regression, k-nearest neighbors, support vector regression, eXtreme Gradient Boosting, and random forests. The performance of these models is assessed based on mean squared error, R² score, and prediction accuracy within 1% and 2% thresholds. Experiment results show that eXtreme Gradient Boosting performed overall with the best accuracy under significant event impact and with a short-term dataset (approximately 2 years). The experiment also takes window rolling to make every model perform more stably. This study discusses the performance of the five models mentioned and uses the designed models to run through the same condition of the dataset to compare their differences. The limitations of the current work are taken into consideration, and directions for future research are identified.