ABSTRACT
The time series are wide-ranging and challenging to forecast in the real world. Since its statistical features fluctuate over time, its distribution likewise varies temporarily, leading to major differences in distribution by current approaches. Long Short-Term Memory (LSTM)+GRU, a modified, full gradient version, is significantly quicker and more accurate. But it is still uncertain how to model the time series from a distribution standpoint. We have used the historical data of the publicly accessible Reliance Industries Limited and Tata Consulting services. We used a data collection of 1750 data points for this purpose. In the end, a comparative analysis of six models for closing stock price was conducted based on SVR, RF, KNN, LSTM, GRU, LSTM+GRU. We also find that the hyper parameters under consideration are nearly independent, and we develop suggestions for their effective modification. All models are hyper tuned by considering several factors such as batch size learning rate, number of Epochs, and model accuracy. LSTM+GRU had the best predictive precision of all these models. As a result, LSTM+GRU is a promising approach for stock market applications that need accurate time interval generation or measurement. The findings will help researchers design more successful models for the forecasting of the stocks.
