Survey of neural network methods for predicting token prices in blockchain markets

Traditional financial models cannot be used as-is for predicting token prices in blockchain markets due to their volatility and fragmented data sources. However, neural network based approaches have been proven to cater for such nonlinear dynamics and thus have the potential for predicting token prices and thus have been also extensively used in this context. In this article, we present a review of these works by organizing the literature around three core areas, i.e ., data inputs, neural architectures and evaluation standards. It synthesizes findings across studies to clarify how researchers use market data, on-chain indicators, sentiment signals and macroeconomic features. We also compare the effectiveness of architectures, such as recurrent models, transformers and graph-based networks. The article identifies common methodological flaws, such as data leakage, non-stationarity and reliance on purely statistical metrics. Moreover, we also outline how future work can move toward interpretable models, benchmark datasets, finance-aware evaluation criteria and more reproducible designs. The primary objective of the article is to provide a clearer foundation for advancing research in neural network methods for predicting token prices so that it is both academically rigorous and practically relevant.