Predicting the stock market index using GRU for the year 2020

Predicting the behavior of the stock exchange is crucial, since it ranks as one of the foremost frequently studied fields nowadays. It is challenging to forecast the financial sector that requires a careful analysis of historical trends. The precision of conventional methods for predicting changes in stock prices was insufficient. In order to anticipate the values of stocks, deep learning becomes particularly popular technique. Deep Learning is a particularly popular method for anticipating stock prices because of the efficiency in patterns that are not linear. Recurrent neural networks were used to increase prediction accuracy since share price movements are dependent on sequence of time. Within Recurrent Neural Networks (RNNs), their pivotal strength lies in their adeptness at managing sequential data and grasping offered a distinct challenge as well as a chance to investigate sophisticated forecasting algorithms. Nonlinear stock price movements cannot be accurately predicted by traditional data analytics approaches. Owing to its effectiveness in capturing intricate, nonlinear patterns within time-series data, deep learning stands out as the predominant method employed for predicting stock prices. A 25% drop in global stock prices and a 30% reduction in G20 countries market prices were recorded in March 2020. Many businesses experience significant losses, and it may be very challenging for them to recover. While many stockholders suffered losses, several businesses managed to continue making money throughout this period. whether the Gated Recurrent Unit (GRU) model, exhibits superior predictive capabilities for significant fluctuations in stock prices during the year 2020 compared to the LSTM model. This research involves a comparative analysis between GRU and LSTM using diverse stock indices, with historical data sourced from the Yahoo Finance API. Notably, the Mean Absolute Percentage Error (MAPE) values demonstrate that GRU outperforms LSTM in forecasting accuracy. The MAPE values for our proposed approach are 1.09, 1.36, 1.43, 0.89, 1.52, and 1.76 for NSE, BSE, Dow Jones, NYSE, S&P, and NASDAQ form January 2015 to March 2020.