GRU, AdaGrad, RMSprop, Adam Implementation of GRU and Adam Optimization Method For Stock Price Prediction

Abstract

In terms of their potential, stocks are one of the most profitable investment options today. If done well and right, stocks can be a very profitable investment. However, volatile stock prices make it necessary to predict stock prices to make a profit. Gated Recurrent Unit (GRU) is a method for predicting time series data such as stock prices. The Optimization method is needed to get accurate prediction results. The weight renewal optimization method such as Adam is implemented to obtain the best weight in the Gated Recurrent Unit (GRU) and to find out the best loss function value generated by the Adam optimization method. The GRU-Adam implementation is carried out on two stock data, namely ICBP and YULE. The results of this research are that the ICBP data yields the respective loss function values, namely train loss 0.0016 and validation loss 0.0007. Whereas the YULE data resulted in a train loss value of 0.0051 and a validation loss of 0.0031. The MAPE generated in the ICBP stock data is 0.97%. While the YULE data is 3.00%.