International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 10 | Oct 2025
p-ISSN: 2395-0072
www.irjet.net
Cryptocurrency Price Forecasting Using Artificial Neural Networks: A Theoretical Framework Rukshad Amaria, College of Computing, New Jersey Institute of Technology, Newark, NJ - 07102 , US https://orcid.org/0009-0003-4291-0213 ---------------------------------------------------------------------***----------------------------------------------------------------------Abstract: Cryptocurrency markets operate in a highly volatile, decentralized, and non-stationary environment that challenges traditional econometric methods. This paper presents a theoretical framework for multi-asset cryptocurrency price forecasting using Artificial Neural Networks (ANNs). A feedforward ANN trained with the back-propagation algorithm is formulated to approximate nonlinear dependencies among blockchain, financial, and sentiment features for Bitcoin, Ethereum, and Solana. Twelve mathematical expressions describe the learning process, and five schematic figures illustrate data flow, architecture, and convergence. The study establishes that ANNs can capture nonlinearities more effectively than linear regression or GARCH models. It provides a foundation for future empirical studies integrating decentralized finance analytics and large-scale sentiment information
[2] incorporated sentiment embeddings with temporal attention to capture emotional market cycles. Aggarwal et al. [3] studied cross-market sentiment links between equities and crypto assets. Zhang and Dagli [4] explored weekday anomalies using ANN classifiers. Li and Gupta [5] compared GARCH and recurrent models for volatility. Hybrid CNN-LSTM designs [6] and transformer-based sentiment fusion [7] improved empirical accuracy but lacked explicit theoretical reasoning. The present paper emphasizes analytical formulation and structural explanation for multi-asset ANN systems.
III. THEORETICAL WORK A. Neuron Computation Each neuron
Keywords: Cryptocurrency, Artificial Neural Network, Backpropagation, Forecasting, Blockchain Analytics, Computational Finance
I. INTRODUCTION Cryptocurrency markets function continuously and display complex price dynamics influenced by global liquidity, mining activity, and investor sentiment. Linear econometric models such as ARIMA and GARCH assume stationarity that does not exist in these markets.
computes an activation
Artificial Neural Networks offer a non-parametric approach to approximate nonlinear mappings from heterogeneous inputs to price outputs. The backpropagation algorithm iteratively adjusts weights to minimize prediction error and adapt to market variation. B. Back Propagation
This paper develops a concise theoretical structure for ANN-based cryptocurrency forecasting. The focus is on mathematical clarity, feature formulation, and generalized learning behavior in decentralized contexts.
For N training examples, the mean-squared error is
II. RELATED WORK Althelaya et al. [1] compared multivariate recurrent architectures for cryptocurrency prediction. Chen et al. © 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 534