Author(s) :
S. S. Akhila, Naveena A Priyadharsini
Conference Name :
International Conference on Modern Trends in Engineering and Management (ICMTEM-25)
Abstract :
Accurate stock price prediction is essential for investors and financial analysts to make informed decisions in volatile markets. Traditional models struggle with nonlinear patterns, necessitating advanced Deep Learning (DL) techniques to improve forecasting accuracy. Stock price prediction is a crucial yet challenging task due to the highly volatile and nonlinear nature of financial markets. This paper presents an advanced DL approach to enhance stock price prediction accuracy by leveraging a hybrid Bidirectional Long Short-Term Memory (BiLSTM) – Gated Recurrent Unit (GRU) model integrated with hierarchical frequency decomposition techniques. The proposed framework begins with historical stock data preprocessing, including handling missing values and data cleaning, followed by feature engineering to extract relevant market indicators. The dataset is then split into training and testing sets, ensuring standardized scaling and normalization to prevent model bias and outdated data issues. The classification model, based on a hybrid BiLSTM-GRU architecture, is employed to capture temporal dependencies and complex stock market patterns effectively. By incorporating hierarchical frequency decomposition, the model further enhances feature extraction, leading to improved predictive performance. Overall, this work is implemented using Python Jupyter software and achieves the predicted error value, such as the Root Mean Square Error (RMSE) of 6.1175 and the Mean Absolute Error (MAE) of 2.8093. This work significantly improves accuracy with the investor and customer behavior analysis of stock performance.