Modeling and Forecasting Stock Prices Using LSTM and XGBoost: A Comparative Machine Learning Perspective

Because financial markets are so volatile, forecasting stock prices is a difficult and everevolving process. Because of advancements in machine learning and deep learning, researchers are now using more than one model to improve how well predictions are established. This essay examines how two celebrated algorithms, XGBoost and LSTM networks, both achieve in stock price forecasting based on history data markets. Both training and testing of the models utilize the same pre-processing and employ Root Mean Squared RMSE (Standard Error), Mean Absolute Error (MAE) and Rsquared ($\mathbf{R}^{\mathbf{2}}$) to measure performance. LSTM which is good at capturing the way events are connected over time, is placed to the test using XGBoost, which is a highly effective and popular ensemble method. The results account for why one method be better than the rest in predicting financial time series and why each of them is better suited to its task purpose. The comparison shows some real-world positive aspects of the models. LSTM recognized long term dependencies and trends better than the other models because of its memory structure, while XGBoost could better manage structured data well and provides fast and interpretable results. The study incorporates short-term future forecasting to evaluate if the models are generalizable beyond the test set.