Comparison of MDKA Stock Price Prediction using Multi-Layer Perceptron, Long Short-Term Memory, and Gated Recurrent Unit

Abstract

Shares are rights owned by a person against a company due to the delivery of capital, either in part or in whole. Investors invest in stocks and try to get maximum results, but many investors are still unsure about the risks involved in investing. To minimize risk, investors need to predict stock prices with an accurate method. Several methods that can be implemented to predict stock data include Multi-Layer Perceptron (MLP), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). The research objective to be achieved in this study is to compare the performance of each algorithm in producing a more accurate stock price forecasting model by testing neurons (10, 20, 30) and epochs (50, 75, 100). The research was conducted on the stock price data of PT. Merdeka Copper Gold Tbk (MDKA) which is a mining sector share with the largest capitalization value. Tests on some of the algorithms above got the best results using 82% training data and 18% test data, namely the MLP model with 10 neurons and 100 epochs with a MAPE training data result of 2.325 and a MAPE test data of 2.014. Based on the test results, MLP can predict MDKA stock prices for the 2018-2022 period with good performance and a relatively small error rate, while tests using the LSTM and GRU methods still produce large errors. Thus, it can be concluded that MLP can predict stock prices with more accurate results.