Multivariate time series stock price data prediction in the banking sector in Indonesia using Bidirectional Long Short-Term Memory (biLSTM)

Abstract

The capital market is a place for individuals or business entities to carry out investment activities, especially in the banking sector, one of the sectors in the LQ45 stock index which is in great demand by investors in Indonesia. In the capital market, one of the investments that can be made is stock investment, but investors will be faced with uncertainty by fluctuations in stock prices caused by several factors, one of which is macroeconomic factors. Therefore, a predictive analysis of stock prices is needed to prevent uncertainty and minimize losses. Accurate prediction models can use deep learning algorithm methods. In the prediction of stock price movements, the data used is historical data on stock prices which is time series type data. This study conducted stock price predictions using the Bidirectional Long Short-Term Memory (biLSTM) method. biLSTM is another variation of the LSTM model. The object of this study uses the variables open, close, adj close, low, high, volume, value, buying rate, selling rate. The data that has been obtained will be preprocessing. Next build a prediction model using hyperparameter tuning with Genetic Algorithm (GA), train the model and evaluate the model. Data testing was carried out using Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) with 4 data from the banking sector in Indonesia including Bank BRI, Bank BNI, Bank BCA, and Bank Mandiri. Based on the data testing that has been carried out, the results of the biLSTM algorithm can predict stock prices accurately because it has a relatively low RMSE value with a MAPE value below 10%.