Predicting Stock Closing Prices: A Deep Learning Approach to Financial Forecasting

In an era of volatile financial markets, accurate stock price prediction is essential for informed investment decisions. This project tackles the challenges of forecasting stock closing prices for TCS.NS, ADANIENT.NS, and ITC.NS using advanced deep learning models over a comprehensive 10-year dataset. Through innovative preprocessing methods—such as forward filling for missing values, normalization for model stability, and reshaping data into sequences for recurrent neural networks (RNNs)—the study employed three architectures: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Convolutional Neural Network (CNN). An ensemble approach that combined predictions from these models was explored to enhance overall prediction accuracy. However, the Gated Recurrent Unit (GRU) model consistently demonstrated superior performance, achieving the highest <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex> score of 0.998185 and the lowest Mean Absolute Percentage Error (MAPE) of 4.68947 in Dataset 1, along with strong results in Dataset 3 with an <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{R}^{\mathbf{2}}$</tex> score of 0.994933. Although the LSTM model showed promise in Dataset-2, it exhibited a significantly higher MAPE, indicating its limitations. The findings underscore the effectiveness of the GRU model for stock price prediction, highlighting the importance of selecting appropriate modeling techniques based on dataset characteristics while establishing a robust framework for financial forecasting using advanced deep learning methodologies.