Benchmarking Modeling Architectures for Cryptocurrency Price prediction using Financial and Social Media Data

The volatility of cryptocurrencies necessitates reliable short-term price prediction models for informed investment decisions. This work presents two benchmarking studies that predict cryptocurrency price over hourly and daily time horizons based on market indicators and social media data. The studies evaluate various models, network architectures, and feature combinations, with a focus on understanding the influence of lag periods, data aggregation, and sentiment analysis nuances on cryptocurrency price. Empirical evaluation identifies LSTM to be the leading prediction model, outperforming other singular models including CNN, ARIMA, and SVR. Hybrid architectures combining LSTM with statistical methods, ARIMA or ARIMAX, demonstrate robust performance over a range of model configurations and combinations. Overall, the findings of this study offer practical contributions on how statistical methods combined with deep learning can address non-stationarity, preprocess social media data, and help support feature interpretability for cryptocurrency price prediction.