Time Step configuration on Long Short-Term Memory Model for Taiwanese Stock Forecast

This code is the implementation of the paper An Effective Way for Taiwanese Stock Price prediction: Boosting the performance with Machine Learning Techniques

Please cite this articale, if the code can help you.

Nguyen HT, Tran TB, Bui PHD. An effective way for Taiwanese stock price prediction: Boosting the performance with machine learning techniques. Concurrency Computat Pract Exper. 2021;e6437. https://doi.org/10.1002/cpe.6437

or

@article{Nguyen2021,
  doi = {10.1002/cpe.6437},
  url = {https://doi.org/10.1002/cpe.6437},
  year = {2021},
  month = jun,
  publisher = {Wiley},
  author = {Hai T. Nguyen and Toan B. Tran and Phuong H. D. Bui},
  title = {An effective way for Taiwanese stock price prediction: Boosting the performance with machine learning techniques},
  journal = {Concurrency and Computation: Practice and Experience}
}

Installation

Python version 3.6 and install the dependencies:

pip install -r requirements.txt

Usage

We implemented Random Forest, Support Vector Regressor, Autoregression, and LSTM for forecasting the stock price.

python stock.py -b 128 -e 200 -m lstm -y 9 -t 1

The explanation of the parameters is explained as follow:

-b: the number of batch size, default value is 128
-e: the number of epochs, default value is 200
-m: model type, default is *lstm* - LSTM, the others are: *rf* - Random Forest, *svm* - Support Vector Regressor, and *autoreg* - Autoregression
-y: training years, the default value is 9, our erexperiments inspected on 9, 10, and 11.
-t: timesteps, the number of timestep for LSTM, the default value is 1.