This work aims to conduct an investigation on 2-, 5- and 10 –output-step with 5 fixed input-step Bitcoin price prediction, using gated recurrent unit (GRU) and long short-term memory (LSTM). The effects of using 2 and 6 layers with regard to LSTM and GRU (2L- and 6L- LSTM and GRU) on the prediction performance are also examined. Two datasets with statistically distinct features, e.g., rather monotonic and non-monotonic, consecutively referred to Binance and Poloniex, the world's leading crypto trading and cryptocurrency exchange platforms are experimented for intensifying the investigation. Prediction performance evaluations include root mean square error (RMSE) and mean absolute error (MAE) along with Pearson correlation coefficient (Corr) are employed here. The best averaged results of all the measures are generated by 2L-GRU. 0.9873, 0.9777 and 0.9593 Corr means are generated by 2-, 5- and 10- output-step; and 0.9758, 0.9575 and 0.9259 Corr means are resulted by the same numbers of steps, respectively for Binance and Poloniex. Overall prediction performance based on more-simpler, monotonic Binance data is rather better than more – complicate, non-monotonic Poloniex data.
Muhammad Shahzeb Khan, Sibghat Ullah Bazai, Muhammad Imran Ghafoor, Shah Marjan · 6 authors
This paper investigates the potential of using a gated recurrent unit (GRU) neural network (NN) for forecasting the prices of three popular cryptocurrencies: Bitcoin (BTC), Ethereum (ETH), and Litecoin (LTC). A dataset spanning from October 2021 to October 2022 was collected and used to train and evaluate the performance of the proposed model. The proposed GRU model was evaluated using the root mean squared error (RMSE) and the mean absolute percentage error (MAPE) as evaluation metrics. The results of the study show that the GRU model achieved an RMSE of 366.0601 and a MAPE of 1.7268% for BTC, an RMSE of 37.6678 and a MAPE of 2.3342% for ETH, and an RMSE of 1.0902 and a MAPE of 1.7278% for LTC. The results indicate that the GRU model performed well in forecasting cryptocurrency prices and holds promise as an approach for further research in this field.
In recent years, the digital world is fast speeding developed from decentralised concept to blockchain, then to cryptocurrency. Especially, cryptocurrency is a popular trending in recent decades that attracts different experts from various field. Its high volatility has been attracted plenty of investors while also brings the difficulty for realizing the price forecasting. On this basis, this study uses public cryptocurrency dataset and three analytical models to predict the direction of cryptocurrency’s price. To be specific, three underlying assets covering large proportion in cryptocurrency are selected, i.e., Bitcoin, Ethereum and Dogecoin. According to the analysis, the prediction results of different models and approaches will be presented. At the end of study, it gains that the optional model with appropriate hyperparameters based on the judgement of metrics values, which offers relevant suggestions for future works. These results shed light on guiding further exploration of cryptocurrency price prediction in terms the state-of-art machine learning scenarios.
Recently, financial globalization is an extremely improved in distinct manners for enhancing service quality with advanced resources. An effectual application of bitcoin Blockchain (BC) approaches allows the shareholders to be concern regarding the risk and return of financial product. The shareholders mainly concentrate on the predictive of risk and return rates of financial product. Thus, an automated return rate bitcoin predictive method develops vital for BC financial product (FP). A newly planned machine learning (ML) and deep learning (DL) techniques offers a way for the return rate predictor systems. This work designs a Lion Swarm Optimization with Deep Learning Driven Predictive Model on Blockchain Financial Product Return Rates (LSODL-BFPRR) technique. The projected LSODL-BFPRR technique lies in the effectual forecasting of return rates in the BC financial sector. In the presented LSODL-BFPRR technique, stacked bidirectional gated recurrent unit (SBiGRU) approach was exploited for return rate classification. To modify the hyperparameters based on the SBiGRU approach, the LSO algorithm is used. The LSODL- BFPRR technique exploits Ethereum (ETH) return rate as the target. The experimental outcomes of the LSODL-BFPRR technique are tested using a series of simulations and the results demonstrate the effectual predicting results of the LSODL-BFPRR technique over other ones.
Since Bitcoin was proposed in 2008, it has become a very valuable asset and an important part of many investors’ portfolios. It’s important to both understand Bitcoin mechanics and predict its valuation with the help of the state-of-art machine learning tools. The study develops four different models, including Ordinary Least Squares (OLS) regression model, Random Forest, Light Gradient Boosting Machine (LightGBM), and Long Short-Term Memory (LSTM), to predict the return of Bitcoin and compare the performance of these models. According to the analysis, the daily changes in the high, low, close price of Bitcoin, and close price of Tesla stock, and gold price between yesterday and today are all strongly correlated to the Bitcoin return on tomorrow. The statistical approach, or OLS modeling, has the simplest algorithm whereas the highest accuracy rate. The LightGBM model and LSTM model have lower accuracy rates in order, but still exceed the 50% (random benchmark). The Random Forest model, as another type of decision tree algorithm, has similar prediction results with the LightGBM model but a lower accuracy rate that fails to reach the benchmark. Based on the analysis, multiple factors affect the Bitcoin return, and these results provide an insight for investors to the cryptocurrency market and the macroeconomic environment. It validates the effectiveness of several machine learning algorithms in Bitcoin return forecasting and supports future developments in related fields.
Price prediction of cryptocurrencies is bound to get more opportunities for investors engaged in digital currency-related industries in order to earn more revenue. In the traditional forecasting methods, the problem of the high volatility of bitcoin price needs to be effectively solved, making the forecasting accuracy become low and ineffective. Due to the rapid development of artificial intelligence technology, more and more relevant algorithms were used for cryptocurrency price research. This study would compare and analyze the prediction effect of the ARIMA time-series model, the Random Forest algorithm of machine learning, and the LSTM algorithm of deep learning algorithm on cryptocurrencies price prediction to assist investors in making investment decisions. In this paper, five years of time-series data of Bitcoin, Ether, and Dogecoin is obtained from 2018 to 2022. Then, the training set and testing set are separated with 0.8:0.2 to test ARIMA, Random Forest, and LSTM algorithms. To evaluate the model, the mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and decidability coefficient (R2) are chosen as metrics to measure the prediction of each prediction model precision. Comparing the experimental results, the prediction accuracy of LSTM is better than that of Random Forest, and the prediction accuracy of Random Forest is better than ARIMA model. These results shed light on guiding further exploration of significant to cryptocurrency industry practitioners and visitors.
Currently, cryptocurrency has become a research and investment topic of great concern and attracted considerable attentions in a wide range of fields. Stocks are well known as a form of investment, and there are countless studies on how its price trends. Cryptocurrencies, however, are not easily predicted due to their extremely high volatility. This paper implements the forecasting results by taking three major cryptocurrencies (i.e., Bitcoin, Ethereum, Dogecoin) as examples for the period starting from January 1st, 2020 to May 31st 2022. Four popular prediction models (XGBoost, LightGBM, GARCH, ARIMA) are applied in Python by training models and testing the prediction results. According to the analysis, XGBoost and LightGBM can forecast the future prices for all three cryptocurrencies exactly apart from the turning points when prices rise and drop suddenly. Although the predicting trends of GARCH, ARIMA have differences from the real price, they can forecast well except the unexpected situations such as COVID-19. Overall, relatively reliable and accurate prediction models can be provided for investors to apply and make wise decisions in investments based on the results of this study.
Contemporarily, blockchains and cryptocurrencies have gained their popularity among investors and hedge funder, where both have bright prospects. On this basis, cryptocurrencies have been used in trading more and more with the development of website and computer. In this case, their prices fluctuations do have great significance to the public. This paper chooses three machine learning model (i.e., XGBoost, LightGBM and Linear Model) to predict the price of three cryptocurrencies (i.e., Bitcoin, Dogecoin and Ethereum). To be specific, this study uses the data from 2020-01-01 to 2022-12-07, including close price, open price, high price, low price, and the volume of trading coins. According to the analysis, Linear Model can predict the price best, with well-fitted trend prediction and accurate price prediction. In addition, other models can also have good predictions but they are not better than Linear model. These results can help others to predict the price of cryptocurrencies and have a deep understanding of cryptocurrency and machine learning.
Predicting Bitcoin price is a universal research area as it attains significance in predicting the market way of its rate so that, investors could procure profits. Concurrently, with the evolution of Machine Learning (ML), researchers attempted to use ML based algorithms for forecasting the Bitcoin price. However, these researches have resulted in inefficient prediction due to error rate. For alleviating such pitfalls, this study intends to forecast the Bitcoin price by comparing its deviations pre and post Covid using suitable ML algorithms. To achieve this, the study proposes Auto Regressive Integrated Moving Average (ARIMA) with Optimized Genetic Algorithm (OGA). In this case, ARIMA model is considered as it possess the innate ability in capturing standard temporal reliances which is distinct to time-series data. Further, hyperparameters are selected by GA based on the fitness function. Based on this, hyperparameter tuning is performed which assist to improvise the model performance. For determining if there exists any deviations in Bitcoin price (pre and post Covid), Augmented Dickey Fuller (ADF) test is considered. Further, comparative analysis is regarded in accordance with performance metrics to validate the performance of the proposed system which proves its effectiveness in predicting Bitcoin price.
Bitcoin is sometimes called the new gold, replacing gold as a hedge against inflation, and research on the relationship between bitcoin and gold has important practical implications. This paper first calculates the correlation between bitcoin and gold. By introducing the calculation of dynamic penalty coefficient, the double stock portfolio investment problem is transformed into the single stock purchase investment problem, which greatly reduces the difficulty of feature engineering and model application. In terms of decision model, deep reinforcement learning (PPO algorithm) is used to make quantitative investment decisions, and the expected data in SLTM is taken as the input data of deep reinforcement learning, which is combined with deep reinforcement learning. Compared with machine learning quantitative investment decisions, after a period of learning, the accuracy rate and returns have been substantially improved.
In recent years, Bitcoin has become the most widely used blockchain platform in business and finance. The goal of this work is to find a viable prediction model that incorporates and perhaps improves on a combina-tion of available models. Among the techniques utilized in this paper are exponential smoothing, ARIMA, artificial neural networks (ANNs) models, and prediction combination models. The study's most obvious discovery is that artificial intelligence models improve the results of compound prediction models. The sec-ond key discovery was that a strong combination forecasting model that responds to the multiple fluctua-tions that occur in the bitcoin time series and Error improvement should be used. Based on the results, the prediction accuracy criterion and matching curve-fitting in this work demonstrated that if the residuals of the revised model are white noise, the forecasts are unbiased. Future work investigating robust hybrid model forecasting using fuzzy neural networks would be very interesting.
Bitcoin, the largest cryptocurrency, is extremely volatile and hence needs a better model for its pricing. In the literature, many researchers have studied the effect of data normalization on regression analysis for stock price prediction. How has data normalization affected Bitcoin price prediction? To answer this question, this study analyzed the prediction accuracy of a Legendre polynomial-based neural network optimized by the mutated climb monkey algorithm using nine existing data normalization techniques. A new dual normalization technique was proposed to improve the efficiency of this model. The 10 normalization techniques were evaluated using 15 error metrics using a multi-criteria decision-making (MCDM) approach called technique for order performance by similarity to ideal solution (TOPSIS). The effect of the top three normalization techniques along with the min–max normalization was further studied for Chebyshev, Laguerre, and trigonometric polynomial-based neural networks in three different datasets. The prediction accuracy of the 16 models (each of the four polynomial-based neural networks with four different normalization techniques) was calculated using 15 error metrics. A 16 × 15 TOPSIS analysis was conducted to rank the models. The convergence plot and the ranking of the models indicated that data normalization plays a significant role in the prediction capability of a Bitcoin price predictor. This paper can significantly contribute to the research with a new normalization technique for utilization in varied fields of research. It can also contribute to international finance as a decision-making tool for different investors as well as stakeholders for Bitcoin pricing.
As cryptocurrencies develop, it cannot be denied that crypto prices are volatile. One of the influencing factors is the increasing volume of transactions which attracts the interest of researchers to conduct research in developing coin price predictions from cryptocurrencies. The method, algorithm and amount of data affect the prediction results. In this study, prediction modelling will be carried out using the LSTM method and short-term data. This study will conduct two experiments using the simple LSTM method and utilising multivariate time series with LSTM. The smallest predicted value is obtained using an 80/20 data allocation distribution scenario, input layer LSTM = 360, Epoch = 500, a Solana coin with RMSE = 0.111, R2 = 0.9962. It can be interpreted that short-term data can be used in making predictive models. Still, special attention needs to be paid to the characteristics of the dataset used and the modelling methodology, and it is hoped that the results of this study can be used in further research.
In this paper, we propose a novel approach for forecasting cryptocurrency portfolios, harnessing modified versions of the N-BEATS deep learning architecture, integrated with convolutional network layers, Transformer mechanisms, and the Mish activation function. Our thorough evaluation, featuring an extensive sample size exceeding 4 million portfolio test samples, shows these variations outperforming traditional and other deep learning forecasting methods across various metrics. Particularly noteworthy is our N-BEATS Perceiver model, a Transformer-based variation, which not only delivers superior forecast accuracy but also exhibits a robust risk profile with less downside. Furthermore, the model performs exceptionally well under the TOPSIS method across a broad spectrum of portfolio evaluation parameters, making it a valuable asset for both portfolio selection and risk management in the dynamic cryptocurrency market.
These days, virtual currencies-the term used to describe cryptocurrencies-are more well-known and have aroused the interest of numerous people. As a result of using blockchain technology, which decentralizes banking, daily cryptocurrency trading has become quite popular among observers, investors, customers, and many other groups. However, due to the daily fluctuations in the value of cryptocurrencies, prediction techniques enable stakeholders to look into the future and identify potential threats to their crucial investment operations. The increasing popularity of cryptocurrencies has made price predictions more promising for investors along with researchers. As artificial intelligence (AI) has advanced to such a level, forecasting the price of cryptocurrencies has grown in significance. In this study, we develop an approach based on machine learning and deep learning as a form of AI to forecast the price of various cryptocurrency exchanges, such as Bitcoin (BTC), Ethereum (ETH), BinanceCoin (BNB), and FTX (FTT), based on their various pricing points. The results from the machine learning models demonstrated that linear regression performed better in forecasting all types of cryptocurrency, with a maximum R2 score of 0.9477, 0.9232, 0.9204, and 0.8925 for BTC, ETH, BNB, and FTT, respectively. However, our study found that the Gated Recurrent Unit (GRU), a deep learning-based model, was the most effective algorithm for predicting the prices of all four types of cryptocurrencies. With GRU, we were able to predict the price of ETH with an R2 score as high as 0.9983, and we were also able to predict the prices of BTC, BNB, and FTT with R2 scores of 0.9969, 0.9772, and 0.9873, respectively. The proposed approach demonstrated superior results with minimal prediction errors on estimating the price of all the different crypto exchanges when the outcomes of our research were dealt with those of the existing studies.
Bitcoin is a high-risk asset with a potentially high return. Predicting Bitcoin candlestick, i.e., open, high, low, and close (OHLC) prices, can help investors make trading decisions. This paper aims to explore various data normalization techniques in neural networks to enhance candlestick price prediction. In this study, the sliding window normalization techniques were compared with the whole set normalization techniques for forecasting the daily Bitcoin OHLC prices using two neural network algorithms: Long Short-Term Memory (LSTM) and Gated Recurrent Unit (GRU). The investigated normalization techniques for both the whole set and the sliding window set included z-score normalization, min-max normalization, and relative change normalization. Historical OHLC prices over several days were used to predict the next day's OHLC prices. The results show that the sliding window normalization techniques outperformed the whole set normalization techniques in terms of RMSE and MAPE with the best technique being the GRU algorithm using the sliding window relative change normalization achieving MAPE of 2.25% and RMSE of 870.52.
L. Thanga Mariappan, J. Arun Pandian, V. Dhilip Kumar, Oana Geman · 6 authors
Cryptocurrency has emerged as a well-known significant component with both economic and financial potential in recent years. Unfortunately, Bitcoin acquisition is not simple, due to uneven business and significant rate fluctuations. Traditional approaches to price forecasting have proven incapable of proving adequate data and solutions because prices can now be forecast in real time. We recommended a machine learning-based alternative for a mortgage lender based on highlighted problems in forecasting the price of Bitcoin. The proposed system included a reinforcement learning algorithm for price estimation and forecasting, as well as a blockchain framework for an efficient and secure environment. The proposed prediction, compared to other state-of-the-art strategies in this sector, demonstrated better performance. In this system, the proposed prediction reached improved consistency, in comparison to other systems, with respect to Monero (XMR), Litecoin (LTC), Oryen (ORY), and Bitcoin (BTC).
The cryptocurrency ecosystem has been the centre of discussion on many social media platforms, following its noted volatility and varied opinions. Twitter is rapidly being utilised as a news source and a medium for bitcoin discussion. Our algorithm seeks to use historical prices and sentiment of tweets to forecast the price of Bitcoin. In this study, we develop an end-to-end model that can forecast the sentiment of a set of tweets (using a Bidirectional Encoder Representations from Transformers - based Neural Network Model) and forecast the price of Bitcoin (using Gated Recurrent Unit) using the predicted sentiment and other metrics like historical cryptocurrency price data, tweet volume, a user's following, and whether or not a user is verified. The sentiment prediction gave a Mean Absolute Percentage Error of 9.45%, an average of real-time data, and test data. The mean absolute percent error for the price prediction was 3.6%.
Phumudzo Lloyd Seabe, Claude Rodrigue Bambe Moutsinga, Edson Pindza
Highly accurate cryptocurrency price predictions are of paramount interest to investors and researchers. However, owing to the nonlinearity of the cryptocurrency market, it is difficult to assess the distinct nature of time-series data, resulting in challenges in generating appropriate price predictions. Numerous studies have been conducted on cryptocurrency price prediction using different Deep Learning (DL) based algorithms. This study proposes three types of Recurrent Neural Networks (RNNs): namely, Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), and Bi-Directional LSTM (Bi-LSTM) for exchange rate predictions of three major cryptocurrencies in the world, as measured by their market capitalization—Bitcoin (BTC), Ethereum (ETH), and Litecoin (LTC). The experimental results on the three major cryptocurrencies using both Root Mean Squared Error (RMSE) and the Mean Absolute Percentage Error (MAPE) show that the Bi-LSTM performed better in prediction than LSTM and GRU. Therefore, it can be considered the best algorithm. Bi-LSTM presented the most accurate prediction compared to GRU and LSTM, with MAPE values of 0.036, 0.041, and 0.124 for BTC, LTC, and ETH, respectively. The paper suggests that the prediction models presented in it are accurate in predicting cryptocurrency prices and can be beneficial for investors and traders. Additionally, future research should focus on exploring other factors that may influence cryptocurrency prices, such as social media and trading volumes.
Samuel Gaskin, Rafay Kalim, Kelvin J. Wallace, David Islip · 6 authors
This article addresses the shortcomings of the existing literature regarding cryptocurrency portfolio construction. First, we address the effectiveness of time-series models that capture stylized features. We perform a comparison study on various methods for estimating distributions for asset returns, including normal, historical, and GARCH models within a CVaR setting. The goal of this comparison is to determine the financial benefits of constructing portfolios based on estimated distributions that consider stylized features of crypto return series. Next, we create and compare various prediction models for cryptocurrencies and integrate them with mean-variance optimization to base performance on portfolio management metrics, such as Sharpe ratio and level of diversification, rather than statistical metrics like accuracy and R<sup>2</sup> on which the literature solely focuses. We determine it is unclear which optimization approach (CVaR or Robust MVO) leads to better crypto portfolios, and so, to address this, we compare optimization procedures on out-of-sample data through a thorough cross-validation of hyperparameters for each technique. We then compare the resulting risk-optimal portfolios from each technique. The results show that a CVaR approach with a GARCH simulation and a decision tree prediction model with robust mean-variance optimization yield portfolios of similar risk. We also show that using statistical metrics to evaluate models may not always yield the best financial performance.
The crypto and digital assets ecosystems have attracted investment, regulators, and speculators to their environment. As the blockchain-based framework can reduce transaction costs, generate distributed trust, and enable decentralized platforms, it has become a potential new base for decentralized business models. Previous studies have highlighted the advantages and drawbacks of each platform, such as interest rates, cost concerns, transparency issues, hacking issues, and hazards. Consequently, it is challenging for investors to evaluate the cryptocurrency trading system which determines the optimum exchanges and crucial aspects. Therefore, in order to rank the optimal digital token trading system, this paper develops an evaluation architecture to determine the various token trading systems. The developed architecture integrates fuzzy theory and the best-worst method (BWM) into the decision-making process to assess decision behaviors regarding preference for digital token trading systems in investors in Taiwan. First, this work establishes the views and parameters by modifying the Delphi method based on a literature review and survey. Second, the fuzzy-BWM is applied to obtain the fuzzy weights of the views and parameters. Then, defuzzification and BWM are used to rank the optimal alternatives of the digital token trading systems for investors. The results indicate that the optimal digital token trading system is the decentralized platform, and the critical parameters are gas fees, interest rates, and the mechanism of savings under fuzzy uncertain scenarios. This means that when considering the uncertain and ambiguous characteristics of the expert decision process in digital token trading systems, the evaluation is decentralized and the gas fees are the most important parameter in the digital token investment platform. Academically, the fuzzy BWM-based decision-making architecture can provide corporations and investors with valuable guidance to rank the optimal digital token trading systems based on fuzzy uncertain scenarios. Commercially, the proposed architecture could provide corporations and investors with a useful model to measure the optimal digital token trading system.
Indri Tri Julianto, Dede Kurniadi, Fakhrun Mahda Khoiriyyah
Non-Fungible Tokens (NFTs) experienced a peak of popularity in Indonesia through content created and sold by an account at OpenSea called Ghozali Everyday in early 2022. Ghozali reportedly earned ± Rp. 1.3 billion from the content he has created. This sparked the curiosity of the Indonesian people to imitate what Ghozali Everyday did in the hope of getting similar benefits. The market price of NFTs is the same as stock prices, which will fluctuate depending on the price of the cryptocurrency because these NFTs can generally be purchased with the cryptocurrency, namely Ethereum. This research was conducted to predict the price of NFTs using the Data Mining Prediction Algorithm. Five algorithms are compared to find the best algorithm: Deep Learning, Linear Regression, Neural Networks, Support Vector Machines, and Generalized Linear Model. The methodology used is Knowledge Discovery in Databases. The NFTs price dataset is taken from the page coinmarketcap.com from 16 November 2021 to 16 November 2022. The results show that the best Data Mining Prediction Algorithm is a Neural Network with a value of The lowest Root Mean Square Error (RMSE) compared to other algorithms, namely 83.617 +/- 18.853 (micro average: 85.590 +/- 0.000). After the Neural Network is used in the Dataset, the graph results show no significant difference between the Closing Price and the Predicted Price.
Twitter sentiment has been found to be useful in predicting whether the price of Bitcoin will rise or fall will climb or decline. Modelling market activity and hence emotion in the Bitcoin ecosystem gives insight into Bitcoin price forecasts. We take into account not just the emotion retrieved not just from tweets, but also from the quantity of tweets. With the goal of optimising time window within which expressed emotion becomes a credible predictor of price change, we provide data from research that examined the link among both sentiment and future price at various temporal granularities. We demonstrate in this study that not only can price direction be anticipated, but also the magnitude of price movement with same accuracy, and this is the study's major scientific contribution. Non-Fungible Token (NFT) has gained international interest in recent years as a blockchain-based application. The most prevalent kind of NFT that can be stored on many blockchains is digital art. We did studies on CryptoPunks, the most popular collection on the NFT market, in examine and depict each and every major ethical challenges. We investigated ethical concerns from three perspectives: design, trade transactions, and relevant Twitter topics. Using Python libraries, a Twitter crawler, and sentiment analysis tools, we scraped data from Twitter and performed the analysis and prediction on bitcoin and NFTs.