Time Series Forecasting of Cryptocurrency Prices with Long Short-Term Memory Networks
El-Sayed M. El-Kenawy*1, Abdelaziz A. Abdelhamid2, Abdelhameed Ibrahim3, Marwa M. Eid1,4, Faris H. Rizk5, Ahmed Mohamed Zaki5
1 Department of Communications and Electronics, Delta Higher Institute of Engineering and Technology, Mansoura, 35111, Egypt
2 Computer Science Department, Faculty of Computer and Information Sciences, Ain Shams University, Cairo, 11566, Egypt
3 School of ICT, Faculty of Engineering, Design and Information & Communications Technology (EDICT), Bahrain Polytechnic, PO Box 33349, Isa Town, Bahrain
4 Faculty of Artificial Intelligence, Delta University for Science and Technology, Mansoura 35712, Egypt
5 Computer Science and Intelligent Systems Research Center, Blacksburg 24060, Virginia, USA
Emails: skenawy@ieee.org; abdelaziz@cis.asu.edu.eg; abdelhameed.fawzy@polytechnic.bh; mmm@ieee.org; faris.rizk@jcsis.org; azaki@jcsis.org
Abstract
The rapid evolution of cryptocurrencies has brought transformative changes to the financial landscape. Cryptocurrency prices, characterized by their inherent volatility, pose challenges for precise forecasting. This study introduces a novel approach to cryptocurrency price forecasting, leveraging Long Short-Term Memory (LSTM) networks, known for discerning temporal dependencies within time series data. Motivated to enhance prediction accuracy, this research investigates the effectiveness of LSTM networks in capturing complexities inherent in cryptocurrency price movements. The proposed methodology involves meticulous data collection and preprocessing, utilizing an extensive dataset from Kaggle. This dataset forms the foundation for predictive modeling and facilitates an in-depth analysis of cryptocurrency price dynamics. Exploratory data analysis, including visualization techniques, and a dedicated Time Series Analysis precede the implementation of predictive models, such as LSTM networks. Results and evaluation showcase promising outcomes, emphasizing the models' precision, accuracy, and explanatory power. The Mean Absolute Error (MAE) of 0.0177 underscores the precision achieved in predicting cryptocurrency prices, while the Mean Squared Error (MSE) of 0.00066 and the R² Score of 0.9486 attest to our models' overall accuracy and explanatory power. This research significantly contributes to understanding cryptocurrency forecasting by incorporating LSTM networks, paving the way for advancements in this evolving domain.
Keywords: Cryptocurrency; Long Short-Term Memory (LSTM) networks; Price Forecasting; Time Series Analysis.