Improved LSTM hyperparameters alongside sentiment walk-forward validation for time series prediction
This study aims to address the common issue of biased estimation errors in time series modeling by analyzing the error in locating ideal hyperparameters and defining appropriate validation methods. Specifically, it focuses on predicting the stock price of Bank Rakyat Indonesia using a combination of...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2025-03-01
|
| Series: | Journal of Open Innovation: Technology, Market and Complexity |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S219985312400252X |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850104997209964544 |
|---|---|
| author | Eko Putra Wahyuddin Rezzy Eko Caraka Robert Kurniawan Wahyu Caesarendra Prana Ugiana Gio Bens Pardamean |
| author_facet | Eko Putra Wahyuddin Rezzy Eko Caraka Robert Kurniawan Wahyu Caesarendra Prana Ugiana Gio Bens Pardamean |
| author_sort | Eko Putra Wahyuddin |
| collection | DOAJ |
| description | This study aims to address the common issue of biased estimation errors in time series modeling by analyzing the error in locating ideal hyperparameters and defining appropriate validation methods. Specifically, it focuses on predicting the stock price of Bank Rakyat Indonesia using a combination of historical stock prices, technical indicators, exchange rates, and news sentiment data, while determining the optimal variables for deep learning models. Employing a deep learning-based Long-Short Term Memory (LSTM) model, the study optimizes hyperparameters alongside walk-forward validation for time series prediction. It explores different combinations of variables and adapts the sliding window approach to the context of the data. The results highlight the importance of optimizing hyperparameters and utilizing walk-forward validation for accurate time series prediction. The model incorporating historical stock prices and sentiment scores outperforms others, achieving an RMSE of 96.61 and MAE of 86.97. Incorporating sentiment scores reduces RMSE by 39.55 % compared to models using only historical stock prices, while adding technical indicators does not yield improvement. This study contributes to the field by addressing the issue of biased estimation errors in time series modeling, offering insights into the optimization of hyperparameters and validation methods for accurate predictions. It also underscores the significance of incorporating sentiment analysis alongside historical stock prices for improved forecasting accuracy. |
| format | Article |
| id | doaj-art-939b0cc92e684552a6db6db0127055f8 |
| institution | DOAJ |
| issn | 2199-8531 |
| language | English |
| publishDate | 2025-03-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Journal of Open Innovation: Technology, Market and Complexity |
| spelling | doaj-art-939b0cc92e684552a6db6db0127055f82025-08-20T02:39:12ZengElsevierJournal of Open Innovation: Technology, Market and Complexity2199-85312025-03-0111110045810.1016/j.joitmc.2024.100458Improved LSTM hyperparameters alongside sentiment walk-forward validation for time series predictionEko Putra Wahyuddin0Rezzy Eko Caraka1Robert Kurniawan2Wahyu Caesarendra3Prana Ugiana Gio4Bens Pardamean5Department of Statistical Computing, Politeknik Statistika STIS, Jakarta 13330, Indonesia; Statistics Indonesia (BPS), Jl. Dr Sutomo 6-8, Jakarta, IndonesiaSchool of Economics and Business Telkom University, Bandung 40257, Indonesia; Research Center for Data and Information Sciences, Research Organization for Electronics and Informatics, National Research and Innovation Agency (BRIN), Bandung 40135, Indonesia; Corresponding author at: School of Economics and Business Telkom University, Bandung 40257, IndonesiaDepartment of Statistical Computing, Politeknik Statistika STIS, Jakarta 13330, Indonesia; Statistics Indonesia (BPS), Jl. Dr Sutomo 6-8, Jakarta, IndonesiaFaculty of Integrated Technologies, Universiti Brunei Darussalam, Gadong BE1410, BruneiDepartment of Mathematics, Universitas Sumatera Utara, Medan 20155, IndonesiaBioinformatics and Data Science Research Centre, Bina Nusantara University, DKI Jakarta 11480, Indonesia; Computer Science Department, BINUS Graduate Program Master of Computer Science Program, Bina Nusantara University, DKI Jakarta 11480, IndonesiaThis study aims to address the common issue of biased estimation errors in time series modeling by analyzing the error in locating ideal hyperparameters and defining appropriate validation methods. Specifically, it focuses on predicting the stock price of Bank Rakyat Indonesia using a combination of historical stock prices, technical indicators, exchange rates, and news sentiment data, while determining the optimal variables for deep learning models. Employing a deep learning-based Long-Short Term Memory (LSTM) model, the study optimizes hyperparameters alongside walk-forward validation for time series prediction. It explores different combinations of variables and adapts the sliding window approach to the context of the data. The results highlight the importance of optimizing hyperparameters and utilizing walk-forward validation for accurate time series prediction. The model incorporating historical stock prices and sentiment scores outperforms others, achieving an RMSE of 96.61 and MAE of 86.97. Incorporating sentiment scores reduces RMSE by 39.55 % compared to models using only historical stock prices, while adding technical indicators does not yield improvement. This study contributes to the field by addressing the issue of biased estimation errors in time series modeling, offering insights into the optimization of hyperparameters and validation methods for accurate predictions. It also underscores the significance of incorporating sentiment analysis alongside historical stock prices for improved forecasting accuracy.http://www.sciencedirect.com/science/article/pii/S219985312400252XTime series modelingLSTMStock price predictionValidation methodsSentiment analysis |
| spellingShingle | Eko Putra Wahyuddin Rezzy Eko Caraka Robert Kurniawan Wahyu Caesarendra Prana Ugiana Gio Bens Pardamean Improved LSTM hyperparameters alongside sentiment walk-forward validation for time series prediction Journal of Open Innovation: Technology, Market and Complexity Time series modeling LSTM Stock price prediction Validation methods Sentiment analysis |
| title | Improved LSTM hyperparameters alongside sentiment walk-forward validation for time series prediction |
| title_full | Improved LSTM hyperparameters alongside sentiment walk-forward validation for time series prediction |
| title_fullStr | Improved LSTM hyperparameters alongside sentiment walk-forward validation for time series prediction |
| title_full_unstemmed | Improved LSTM hyperparameters alongside sentiment walk-forward validation for time series prediction |
| title_short | Improved LSTM hyperparameters alongside sentiment walk-forward validation for time series prediction |
| title_sort | improved lstm hyperparameters alongside sentiment walk forward validation for time series prediction |
| topic | Time series modeling LSTM Stock price prediction Validation methods Sentiment analysis |
| url | http://www.sciencedirect.com/science/article/pii/S219985312400252X |
| work_keys_str_mv | AT ekoputrawahyuddin improvedlstmhyperparametersalongsidesentimentwalkforwardvalidationfortimeseriesprediction AT rezzyekocaraka improvedlstmhyperparametersalongsidesentimentwalkforwardvalidationfortimeseriesprediction AT robertkurniawan improvedlstmhyperparametersalongsidesentimentwalkforwardvalidationfortimeseriesprediction AT wahyucaesarendra improvedlstmhyperparametersalongsidesentimentwalkforwardvalidationfortimeseriesprediction AT pranaugianagio improvedlstmhyperparametersalongsidesentimentwalkforwardvalidationfortimeseriesprediction AT benspardamean improvedlstmhyperparametersalongsidesentimentwalkforwardvalidationfortimeseriesprediction |