Accurate stock prices prediction on Grouped Time Series Data using Recurrent Neural Network Variants

Abstract

Stocks are an alluring venture choice since they can create enormous benefits contrasted with different organizations. The development of stock cost designs on the financial exchange is exceptionally unique; consequently it requires precise information demonstrating to gauge stock costs with a low mistake rate. Estimating models utilizing Deep Learning are accepted to have the option to precisely foresee stock cost developments utilizing time-series information, particularly the Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU) calculations. Be that as it may, a few past execution concentrates on have not had the option to acquire persuading precision outcomes. This work given the execution of the estimating technique by arranging the development of time-sequence information on organization stock costs into three gatherings utilizing GRU and LSTM. The precision of the fabricated model is assessed utilizing misfortune elements of Rooted Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE). The outcomes showed that the presentation assessment of the two models are precisely GRU is generally better than LSTM. The most noteworthy approval of GRU was 98.93% of RMSE and 97.78% of MAPE, while the LSTM approval was 94.23% of RMSE and 96.61% of MAPE.