Utilizing Transformer Models and Graph Neural Networks for Timestamp-Based Cryptocurrency Price Prediction: A Deep Learning Approach

No Thumbnail Available
Singh, Aniket
Issue Date
MSc in Business Analytics
Dublin Business School
This study delves into the realm of cryptocurrency price prediction using cutting-edge deep learning techniques, specifically Transformer models and Graph Neural Networks (GNN). We conduct a comprehensive evaluation, benchmarking these methods against traditional models like ARIMA, Simple RNN, and Prophet on Ethereum (ETH) and Bitcoin (BTC) closing prices. Notably, the Transformer model showcases remarkable accuracy in BTC_close predictions, boasting an RMSE of 0.02395 and MAE of 0.02312, surpassing the performance of ARIMA. Conversely, GNN emerges as the top performer for ETH_close, delivering an impressive RMSE of 1111.39 and MAE of 1055.11. Despite its computational simplicity, Simple RNN falls short in comparison. This research contributes valuable insights into harnessing state-of-the-art deep learning architectures for accurate cryptocurrency price forecasting, highlighting the efficacy of Transformer models and GNNs in capturing intricate temporal dependencies within the dynamic cryptocurrency market landscape.