This repository contains the code of our paper entitled " An Exploratory Comparison of LSTM and BiLSTM in Stock Price Prediction". The data we use for the experiment can be downloaded here.

Abstract

Forecasting stock prices is a challenging topic that has been the subject of many studies in the field of finance. Using machine learning techniques, such as deep learning, to model and predict future stock prices is a potential approach. Long Short-Term Memory (LSTM) and Bidirectional Long Short-Term Memory (BiLSTM) are two common deep learning models. The finding of this work is to discover which activation function and which optimization method will influence the performance of the models the most. Also, we implement the comparison of closely related models: vanilla RNN, LSTM, and BiLSTM to discover the best model for stock price prediction. Experimental results indicated that BiLSTM with ReLU and Adam method achieved the best performance in the prediction of stock price.

Implementation

The source code to implement different models for comparison is given in the src folder. To reproduce our result, first you need to clone the repository:

git clone https://github.com/quocviethere/An-Exploratory-Comparison-of-LSTM-and-BiLSTM-in-Stock-Price-Prediction

Then simply run whichever model provided:

python src/bilstm-af.py

Results

Using Adam Optimization and ReLU activation function, we yielded the results as follows:

Model	RMSE	MPE	MAPE	MSE	MAE	$R^2$ score
BiLSTM	3.4927	0.0028	0.0182	12.1992	2.7009	0.8969
LSTM	3.6231	0.0021	0.0198	13.1273	2.9242	0.8891
Vanilla RNN	3.6539	0.0062	0.0197	13.3514	2.9062	0.8872

We further investigate the performance of BiLSTM with different activation functions:

Activation function	RMSE	MPE	MAPE	MSE	MAE	$R^2$ score
ReLU	3.4927	0.0028	0.0182	12.1992	2.7009	0.8969
Tanh	3.5385	0.0038	0.0185	12.5210	2.7445	0.8942
Sigmoid	3.6059	-0.0005	0.0190	13.0031	2.8193	0.8901

and optimization methods:

Optimizer	RMSE	MPE	MAPE	MSE	MAE	$R^2$ score
Adam	3.4927	0.0028	0.0182	12.1992	2.7009	0.8969
RMSprop	3.7757	0.0108	0.0205	14.2565	3.0363	0.8795
SGD	8.6535	0.0255	0.0450	74.8843	6.8549	0.3675

Acknowledgement:

This research is funded by University of Economics Ho Chi Minh City (UEH), Vietnam.

To cite our paper, please use:

@InProceedings{10.1007/978-981-99-5166-6_35,
author="Viet, Nguyen Q.
and Quang, Nguyen N.
and King, Nguyen
and Huu, Dinh T.
and Toan, Nguyen D.
and Thanh, Dang N. H.",
editor="Ranganathan, G.
and Papakostas, George A.
and Rocha, {\'A}lvaro",
title="An Exploratory Comparison of LSTM and BiLSTM in Stock Price Prediction",
booktitle="Inventive Communication and Computational Technologies",
year="2023",
publisher="Springer Nature Singapore",
address="Singapore",
pages="513--524",
isbn="978-981-99-5166-6"
}