Cryptocurrency Price Prediction Using Parallel Gated Recurrent Units
Milad Asadpour, Alireza Rezaee, Farshid Hajati · 2025
Read Cryptocurrency Price Prediction Using Parallel Gated Recurrent Units by Milad Asadpour, Alireza Rezaee, Farshid Hajati (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2512-22599.txt.
Introduction
Milad Asadpour, Alireza Rezaee, and Farshid Hajati forecast cryptocurrency prices with a bank of recurrent networks running in parallel. Bitcoin's blockchain-secured market swings hard, drawing investors who all want foresight. Their model, Parallel Gated Recurrent Units, assigns separate recurrent networks to distinct price-related feature streams, letting each network forecast independently from its own view of the market.
A final neural network then combines the parallel forecasts into one predicted price. The authors report mean absolute percentage errors near three percent for a twenty-step window and under three percent for a fifteen-step window in their tests. They claim higher accuracy and efficiency than current methods while using fewer input data and less computation, a combination aimed squarely at practical traders.
Catalog record
| CREATOR | Milad Asadpour, Alireza Rezaee, Farshid Hajati |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 26 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2512-22599
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Cryptocurrency Price Prediction Using Parallel Gated Recurrent Units Milad Asadpour, Alireza Rezaee, Farshid Hajati Abstract According to the advent of cryptocurrencies and Bitcoin, many investments and businesses are now conducted online through cryptocurrencies. Among them, Bitcoin uses blockchain technology to make transactions secure, transparent, traceable, and immutable. It also exhibits significant price fluctuations and performance, which has attracted substantial attention, especially in financial sectors. Consequently, a wide range of investors and individuals have turned to investing in the cryptocurrency market. One of the most important challenges in economics is price forecasting for future trades. Cryptocurrencies are no exception, and investors are looking for methods to predict prices; various theories and methods have been proposed in this field. This paper presents a new deep model, called Parallel Gated Recurrent Units (PGRU), for cryptocurrency price prediction. In this model, recurrent neural networks forecast prices in a parallel and independent way. The parallel networks utilize different inputs, each representing distinct price-related features. Finally, the outputs of the parallel networks are combined by a neural network to forecast the future price of cryptocurrencies. The experimental results indicate that the proposed model achieves mean absolute percentage errors (MAPE) of 3.243% and 2.641% for window lengths 20 and 15, respectively. Our method therefore attains higher accuracy and efficiency with fewer input data and lower computational cost compared to existing methods. 1 Introduction The current economic system is based on intermediaries and third parties such as banks and financial institutions. These intermediaries are required for most financial transactions, but this reliance introduces drawbacks including limited trust, security, transparency, and flexibility. In 2009, Bitcoin was introduced to the public to facilitate financial transactions and address some of the shortcomings of the current financial system [ 1 ] . Bitcoin does not exist physically; users hold digital accounts identified by public and private keys. Account information, along with transaction data, is recorded in a global ledger (the blockchain). Maintaining this ledger requires substantial computational power to verify and record data across the distributed network. Bitcoin is generated by solving computationally expensive mathematical problems, creating a direct peer-to-peer connection between individuals in each transaction and eliminating intermediaries and third parties. There are no central servers or financial institutions governing transfers. A decade after its advent, Bitcoin has received widespread attention due to its increasing value and volatile performance. From January 2010 to January 2021, the lowest recorded price was $0.1 in July 2010, and the highest was $40,921 in January 2021. Over the same period, Bitcoin’s market capitalization rose from around $16.3 billion in February 2017 to nearly $760 billion in January 2021. Although price movements appear highly volatile and possibly random, closer inspection (e.g., via logarithmic price diagrams) indicates the presence of recurrent patterns [ 2 , 3 ] . These fluctuations and substantial increases in the price of cryptocurrencies have motivated many researchers to search for influential factors and robust methods for predicting Bitcoin prices. Various approaches, including technical and fundamental analysis in economics, have been used for this purpose [ 25 , 5 , 6 ] . Machine-learning methods are among the most promising tools for predicting future cryptocurrency prices. They can generally be categorized into classification and regression approaches. In classification tasks, the goal is to predict a directional movement (e.g., increase or decrease) [ 7 , 8 ] , whereas in regression tasks, the goal is to predict the actual price value. Deep neural networks (DNN), recurrent neural networks (RNN), and convolutional neural networks (CNN) have been widely applied to both regression and classification problems in this domain. Similar advanced deep architectures have been successfully used in other domains such as face analysis [ 35 , 39 , 36 , 37 , 27 , 34 , 26 , 41 , 32 ] , load forecasting [ 31 ] , and medical decision support [ 33 , 38 , 40 , 42 , 56 , 43 , 44 , 45 ] . In this work, we study and predict the price of Bitcoin as a regression problem using a new model called Parallel Recurrent Neural Network (PRNN), instantiated via gated recurrent units (GRUs). Our approach is also motivated by prior work on evolutionary and learning-based optimization in control and communications [ 50 , 46 , 47 , 54 , 51 , 53 , 52 , 60 , 59 , 48 ] and in complex multi-agent and educational systems [ 61 , Rezaee2015coevolutionary ] . The proposed structure comprises three small neural networks. The first and second networks are based on GRU layers and are arranged in parallel, operating independently. The first network predicts the Bitcoin price based on historical price features. The second network predicts the price based on structural features of the Bitcoin network. Finally, the outputs of these two networks are fed into a third network—a feedforward neural network—which combines them based on their respective errors and reliabilities to produce the final price prediction for the next few days. By using different but complementary input data (price and structural features) in parallel networks, price fluctuations and patterns are captured more completely, yielding highly accurate and reliable forecasts. Our experimental results show that the proposed model attains MAPE of 3.243% and 2.641% for window lengths 20 and 15, respectively. Thus, the method uses fewer input features while achieving higher accuracy and lower computational cost compared to existing approaches. 2 Related Work Numerous studies have aimed to identify accurate indicators and influential factors for Bitcoin prices and other digital currencies. Teker et al. [ 9 ] investigated relationships between the prices of gold, crude oil, and major cryptocurrencies such as Bitcoin, Ethereum, and Litecoin. Their findings help explain cryptocurrency price fluctuations at the international level. Peng et al. [ 10 ] combined an autoregressive conditional heteroscedasticity model with support vector regression to predict the prices of Bitcoin and Litecoin, achieving better performance than autoregressive moving-average methods. Karasu et al. [ 11 ] used SVMs with linear and polynomial kernels, trained and evaluated via k -fold cross-validation, and concluded that SVMs generally outperform linear regression. Hitam et al. [ 12 ] further improved SVM performance by optimizing its coefficients with particle swarm optimization (PSO), resulting in more accurate and relatively faster predictors. Radityo et al. [ 13 ] applied genetic algorithms to optimize the weights of a multilayer neural network, helping the model avoid local minima and improving training quality. Rathan et al. [ 14 ] proposed a hybrid approach combining decision trees and linear regression, achieving lower computational cost and slightly higher accuracy than some earlier methods. More recently, researchers have focused on deep learning. Chen et al. [ 7 ] used feature selection to identify twelve critical features affecting the Bitcoin price and then applied classification methods such as long short-term memory (LSTM) networks and quadratic discriminant analysis to predict future price direction (up or down). Their results showed that using these twelve features with LSTMs yields higher accuracy than other classification methods. Wu et al. [ 15 ] proposed an LSTM-based forecasting framework using time-series data of Bitcoin prices. Tandon et al. [ 16 ] used an LSTM network trained with k -fold cross-validation, significantly increasing computational cost but improving the model’s ability to