Research and Analysis of Apple Stock Prediction Based on Improved XGBoost Model

Abstract

Financial markets are known for high volatility and complex non-linear dynamics, so predicting stock prices accurately is always a difficult task for both researchers and practitioners. This study proposes an improved XGBoost model for Apple Inc. (AAPL) stock price prediction. The model uses Bayesian optimization to tune hyperparameters, and it also applies walk-forward validation to keep the time order of data and avoid look-ahead bias. The proposed model is compared with several baseline models, including standard XGBoost, LSTM networks, SVR, and linear regression. The experiments are conducted on a ten-year dataset from Yahoo Finance, and various features such as technical indicators and lagged prices are engineered. The results show that the improved XGBoost achieves an R² score of 95.30%, RMSE of 0.5570, and MAPE of 0.1385, which are better than all the baseline models. SHAP analysis is also used to find that technical indicators and lag features are the most important predictors. These findings suggest that the optimized gradient boosting framework can provide accurate, interpretable, and efficient solutions for real-time financial forecasting and algorithmic trading.