Jong-woo Ryu, Geon-hee Baek, Choon-Sung Nam, Journal of Internet Computing and Services, Vol. 26, No. 3, pp. 111-119, Jun. 2025
10.7472/jksii.2025.26.3.111, Full Text:  HTML
Keywords: Stock Price Prediction, LSTM, Features, Unidirectional, Bidirectional, RMSE

Abstract

This paper analyzes the effect of whether or not to utilize additional features on th model performance in stock price prediction using the LSTM(Long Short-Term Memory) model and compares the performance difference between the unidirectional and bidirectional models. Opening price, closing price, low price, high price, and trading volume were used as input values, and the RSI(Relative Strength Index) and volatility were selected as additional features. The performance of the unidirectional and bidirectional LSTM models that did not include features and the unidirectional and bidirectional LSTM models were compared and analyzed, respectively, and the performance evaluation was evaluated based on the RMSE(Root Mean Square Error). As a result of the experiment, it was confirmed that the model to which the RSI and volatility were added showed a lower RMSE in the prediction performance and improved accuracy. In addition, the bidirectional LSTM model showed better overall performance than the unidirectional LSTM model.

Statistics
Cite this article