System analysis and modelling tools for cryptocurrency price forecasting

Abstract

In the development of financial technology, cryptocurrencies present a complex challenge for price forecasting due to their volatile nature, influenced by various factors from market sentiment to regulatory changes. This paper examines the application of System Analysis and Machine Learning (ML) tools to enhance the accuracy of cryptocurrency price predictions. We review the literature on econometric and ML models, highlighting their potential and limitations in the context of financial markets. The study systematically assesses tools such as ARIMA and Prophet, deep learning frameworks like Keras with TensorFlow and PyTorch, and advanced neural network architectures including LSTM, Transformer, and Temporal Fusion Transformer. The findings reveal that traditional models like ARIMA offer ease of use but need help with the non-linear patterns inherent to cryptocurrency data. In contrast, models like the Temporal Fusion Transformer provide high accuracy but require substantial training time. This research suggests a synergistic approach, integrating ML predictions with system analysis to enhance forecasting accuracy. It proposes the exploration of data clustering by periods and asset types as a promising direction for future research. This composite methodology holds the potential to significantly improve economic forecasting and asset management in the cryptocurrency domain.