A HYBRID DEEP LEARNING AND STATISTICAL APPROACH FOR STOCK PRICE FORECASTING USING WAVELET DECOMPOSITION AND XGBOOST

Analyzing and predicting the stock price movements accurately has been a critical challenge because of the inherent non-linearity, noise, and volatility present in the data.In this project, we proposed a hybrid forecasting framework that combines statistical modeling with deep learning and signal processing techniques.In the proposed approach, we integrated ARIMA to determine linear trend extraction, wavelet decomposition for denoising and separating the high frequency components, and a deep learning architecture which consists of Convolution Neural Network (CNN), Bidirectional Long Short-Term Memory (Bi-LSTM), and an attention mechanism for complex temporal dependencies.To increase the accuracy, we further added XGBoost as an ensemble regressor to combine the outputs of ARIMA and Deep Learning modules.In this, we take the historical data using Yahoo Finance via the ticker symbol of the company.The historical data of Reliance Industries Ltd. is taken from Yahoo Finance for the period from 1 January 2022 to 31 December 2024 to determine model accuracy.The proposed hybrid model achieved higher performance among key evaluation metrics like MSE, MAE, RMSE, MAPE, R-squared, and Directional accuracy, which shows the robustness and effectiveness of the hybrid model in capturing both short-term fluctuations and long-term trends in stock prices.