Stock Price Prediction Using RNNs: A Comparative Analysis of RNN, LSTM, GRU, and BiRNN

Stock price prediction is a crucial area of financial market research, having significant implications for investors, traders, and analysts. However, given the dynamic and intricate nature of financial markets—which are impacted by a wide range of variables such as economic statistics, geopolitical developments, and market sentiment—accurately projecting stock prices is intrinsically difficult. Conventional techniques frequently fail to fully capture these dynamics, producing predictions that are not ideal. Recurrent Neural Networks (RNNs), one of the most recent developments in machine learning, provide potential methods to overcome these obstacles. Despite their potential, the effectiveness of different RNN architectures in stock price prediction remains an area of active research. This study compares four Recurrent Neural Network (RNN) architectures—Simple RNN, Long-Short Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bidirectional Recurrent Neural Network (BiRNN)—for forecasting the Nifty 50 index values on the Indian National Stock Exchange (NSE) from the year 2000 to 2021. Using a comprehensive dataset spanning two decades, we assess each model's performance using the metrics Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Mean Squared Error (MSE). The data shows that the BiRNN model regularly outperforms the other models in all criteria i.e., MAE, MAPE, and MSE, indicating higher predictive accuracy. This study adds to the existing research by offering useful insights into the usefulness of RNN models, especially that of the BiRNN model for predicting stock prices, specifically in the setting of the Nifty 50 index. Our findings emphasize BiRNN's potential as a stock price prediction model and open new options for future research in this area.