A Deep Learning Approach to NIFTY 100 Stock Price Prediction Using LSTM and GRU Networks

Given the fact that financial time series data is inherently random, it is an important task to forecast stock market. In this work, a mixed model is developed using LSTM and GRU to forecast the Close price of the NIFTY 100 share index. The model is fitted with historical stock price data of year 2007 to 2023 with features consisting of volume, open, high, low, Relative Strength Index (RSI) and Simple Moving Average (SMA). Using ‘Trend’ and ‘Seasonal’ terms this hybrid model not only extractslong term trend and the oscillations part of the data but helpsto build a better model for stock price prediction. The modelwas assessed based on its predictive outcome measures such as Mean Absolute Error (MAE), Mean Squared Error (MSE)and accuracy measures. From our tests, MSE was 15227.7711, MAE 101.4199, and accuracy was 97.44% which proved that hybrid LSTM-GRU model can predict stock prices accurately. When plotted, the residuals and the model's predictions against actual prices, the maps reinforced the model's generalization's strength even during extreme market fluctuations. The noveltyof this study lies in the unique combination of LSTM and GRU layers, trained using vast data of nifty 100 as inputs, which involves highly dynamical stock fluctuations; it also includes technical indicators that cannot be considered by conventional RNN models. the novelty of the paper also lies in the predictive analysis done on nifty 100 as, unlike other indexes like nifty 50, this index has not been explored for predictive analysis using machine learning yet. Surprisingly this approach yields high levels of predictability low levels of error rates and hence should provide a reliable method of predicting stock prices in volatile financial environments.