Comparative Analysis of Deep Learning Models for Stock Price Prediction in the Indian Equity Market (BSE)

Abstract

Purpose : This study analyzed and compared the performance of deep learning models for predicting stock prices. Inaccurate prediction of stock prices may lead to misallocation of funds in an investment portfolio, which prompted us to derive actionable insights using deep learning models for investment decision-making.

Methodology : The daily closing price time series data of the BSE SENSEX 50 based on market capitalization as of July 31, 2025, were sourced from Yahoo Finance from August 1, 2015, to July 31, 2025. The data were preprocessed, and models such as recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU) were trained for prediction using Python in the Google Colab environment. Evaluation metrics were employed to evaluate the performance of the deep learning models.

Findings : We found that the GRU was the best-performing model among the three deep learning architectures, followed by LSTM and RNN. However, the deep learning models did not outperform the naive benchmark, which achieved equal or lower prediction errors for 44 of the 45 stocks.

Practical Implications : It is recommended that investors consider stock-specific prediction performance and evaluate predictions from multiple deep learning models and performance metrics before incorporating model-based predictions for investment.

Originality : Unlike prior research on predicting stock prices, the current work compared deep learning techniques by integrating a multi-metric evaluation of companies, ensuring that results are not just technically valid but also practically relevant to market participants.