<abstract><p>In this paper, a distributed machine-learning strategy, i.e., federated learning (FL), is used to enable the artificial intelligence (AI) model to be trained on dispersed data sources. The paper is specifically meant to forecast cryptocurrency prices, where a long short-term memory (LSTM)-based FL network is used. The proposed framework, i.e., <italic>F-LSTM</italic> utilizes FL, due to which different devices are trained on distributed databases that protect the user privacy. Sensitive data is protected by staying private and secure by sharing only model parameters (weights) with the central server. To assess the effectiveness of <italic>F-LSTM</italic>, we ran different empirical simulations. Our findings demonstrate that <italic>F-LSTM</italic> outperforms conventional approaches and machine learning techniques by achieving a loss minimal of $ 2.3 \times 10^{-4} $. Furthermore, the <italic>F-LSTM</italic> uses substantially less memory and roughly half the CPU compared to a solely centralized approach. In comparison to a centralized model, the <italic>F-LSTM</italic> requires significantly less time for training and computing. The use of both FL and LSTM networks is responsible for the higher performance of our suggested model (<italic>F-LSTM</italic>). In terms of data privacy and accuracy, <italic>F-LSTM</italic> addresses the shortcomings of conventional approaches and machine learning models, and it has the potential to transform the field of cryptocurrency price prediction.</p></abstract>
P. V. Nagamani, Gowri Anand, Srinivasa Prasanna, Basava Raju · 5 authors
The past several years have seen an increase in interest in trading that is supported by machine learning and artificial intelligence.Utilize automated trading with the aid of machine learning and artificial intelligence to reap the maximum rewards from the cryptocurrency market.For a specific time, we keep the daily data.We achieve excellent results by utilising tactics supported by cutting-edge algorithms.The results produced the expansion in the crypto currency industry with the aid of straight forward architecture and algorithms.The rise in market capitalization has led to a rise in popularity for the cryptocurrency in 2017.Today's market involves more than 1500 crypto currencies.For usage in online transactions, the crypto currency can be created.A crypto money technology is bitcoin.Bitcoin's value changes constantly, second by second.As a result, we apply machine learning architecture to forecast the value of the bitcoin price in this case.We are working to demonstrate that, in comparison to previous techniques and architectures, this ML architecture produces results that are more accurate.Our study use the Support Vector Machine(SVM) and K Nearest Neighbor(KNN)algorithms to successfully forecast bitcoin prices.The findings demonstrate that the Support Vector Machine(SVM) method outperforms the K Nearest Neighbor(KNN) method as it is currently being used.
The volatile and unpredictable nature of the cryptocurrency market makes it particularly challenging to make profitable investment decisions. different machine learning-based techniques have been employed for forecasting cryptocurrency value. However, although some works have addressed incorporating the Blockchain transactions’ data into the analysis, none of them has provided a hybrid solution, including features obtained through complex network modeling. In this paper, we investigated the use of machine learning and complex network techniques to improve the profitability of a cryptocurrency portfolio during a downtrend period. We extracted features through a complex network-building methodology based on the Bitcoin blockchain transactions, merged them with the historical cryptocurrency values, and generated the predictions using different machine-learning models. The results indicated that incorporating complex network features improved the performance in retaining the initial capital at the end of the experiment, leading to an increment of 7.09% and 4.33% for the CNN and LSTM models, respectively. Our findings suggest that the proposed method enhanced the performance of cryptocurrency investment strategies during downtrend periods.
Faheem Aslam, Zil-e-huma, Rashida Bibi, Paulo Ferreira
We take the novel Twitter-based economic uncertainty (TEU) to examine if it has cross-correlation characteristics with four major cryptocurrencies i.e. Bitcoin, Ethereum, Litecoin, and Ripple. To conduct a more thorough analysis, we apply multifractal detrended cross-correlation analysis (MFDCCA) on seasonal-trend decomposition using Loess (STL) decomposed series as well as without decomposed series on the daily data, ranging from 1 June 2011 to 30 June 2021. The findings of this study indicate that: (i) all pairs of TEU with cryptocurrencies are multifractal and have power-law behavior; (ii) the pairs of Ethereum and Bitcoin with TEU are found to be the most multifractal while Litecoin with TEU has the lowest multifractal characteristics; (iii) all STL decomposed series of cryptocurrency have persistent cross-correlation with TEU with the exception of Ethereum which has anti-persistent cross-correlation with TEU; (iv) all without decomposed series of cryptocurrencies show significant persistent cross-correlation characteristics with TEU; (v) the highest linkage is found for the pair of Bitcoin with TEU. Moreover, to reveal the dynamic characteristics in the cross-correlation of TEU with cryptocurrencies, the rolling window is employed for MFDCCA. These findings have important managerial and academic implications for policymakers, investors, and market participants.
<abstract> <p>This research attempts to fit a polynomial auto regression (PAR) model to intraday price data of four major cryptocurrencies and convert the model into a real-time profitable automated trading system. A PAR model was constructed to fit cryptocurrencies' behavior and to attempt to predict their short-term trends and trade them profitably. We used machine learning (ML) procedures enabling our system to train using minutes' data for six months and perform actual trading and reporting for the next six months. Results have shown that our system has dramatically outperformed the naive buy and hold (B &amp; H) strategy for all four examined cryptocurrencies. Results show that our system's best performances were achieved trading Ethereum and Bitcoin and worse trading Cardano. The highest net profit (NP) for Bitcoin trades was 15.58%, achieved by using 67 minutes bars to form the prediction model, compared to −44.8% for the B &amp; H strategy. Trading Ethereum, the system generated 16.98% NP, compared to −33.6% for the B &amp; H strategy, 61 minutes bars. Moreover, the highest NPs achieved trading Binance Coin (BNB) and Cardano were 9.33% and 4.26%, compared to 0.28% and −41.8% for the B &amp; H strategy, respectively. Furthermore, the system better predicted Ethereum and Cardano uptrends than downtrends while it better predicted Bitcoin and BNB downtrends than uptrends.</p> </abstract>
Cryptocurrency price prediction has garnered significant attention due to the growing importance of digital assets in the financial landscape. This paper presents a comprehensive study on predicting future cryptocurrency prices using machine learning algorithms. Open-source historical data from various cryptocurrency exchanges is utilized. Interpolation techniques are employed to handle missing data, ensuring the completeness and reliability of the dataset. Four technical indicators are selected as features for prediction. The study explores the application of five machine learning algorithms to capture the complex patterns in the highly volatile cryptocurrency market. The findings demonstrate the strengths and limitations of the different approaches, highlighting the significance of feature engineering and algorithm selection in achieving accurate cryptocurrency price predictions. The research contributes valuable insights into the dynamic and rapidly evolving field of cryptocurrency price prediction, assisting investors and traders in making informed decisions amidst the challenges posed by the cryptocurrency market.
As the adoption of cryptocurrencies, especially Bitcoin (BTC) continues to rise in today’s digital economy, understanding their unpredictable nature becomes increasingly critical. This research paper addresses this need by investigating the volatile nature of the cryptocurrency market, mainly focusing on Bitcoin trend prediction utilizing on-chain data and whale-alert tweets. By employing a Q-learning algorithm, a type of reinforcement learning, we analyze variables such as transaction volume, network activity, and significant Bitcoin transactions highlighted in whale-alert tweets. Our findings indicate that the algorithm effectively predicts Bitcoin trends when integrating on-chain and Twitter data. Consequently, this study offers valuable insights that could potentially guide investors in informed Bitcoin investment decisions, thereby playing a pivotal role in the realm of cryptocurrency risk management.
The rapid expansion of the non-fungible token (NFT) market has attracted many investors. However, studies on the NFT price fluctuations have been relatively limited. To date, the machine learning approach has not been used to demonstrate a specific error in NFT sale price fluctuation prediction. The aim of this study was to develop a prediction model for NFT price fluctuations using the NFT trading information obtained from OpenSea, the world’s largest NFT marketplace. We used Python programs to collect data and summarized them as: NFT information, collection information, and related account information. AdaBoost and Random Forest (RF) algorithms were employed to predict the sale price and price fluctuation of NFTs using regression and classification models, respectively. We found that the NFT related account information, especially the number of favorites and activity status of creators, confer a good predictive power to both the models. AdaBoost in the regression model had more accurate predictions, the root mean square error (RMSE) in predicting NFT sale price was 0.047. In predicting NFT sale price fluctuations, RF performed better, which the area under the curve (AUC) reached 0.956. We suggest that investors should pay more attention to the information of NFT creators. We anticipate that these prediction models will reduce the number of investment failures for the investors.
Recently, reinforcement learning has been applied to cryptocurrencies to make profitable trades. However, cryptocurrency trading is a very challenging task due to the volatility of the market, especially during bearish periods. In addressing this problem, the existing literature employs single-agent techniques such as deep Q-network (DQN), advantage actor-critic (A2C), and proximal policy optimization (PPO), or their ensembles. Moreover, in the context of cryptocurrencies, the mechanisms for restricting losses during a bearish market are insufficiently robust. Consequently, the performance of reinforcement learning methods for cryptocurrency trading in the existing literature is constrained. To overcome this limitation, in this paper, we propose a novel cryptocurrency trading method based on multi-agent proximal policy optimization (MAPPO) with a collaborative multi-agent scheme and a local-global reward function to optimize both the individual and collective performance of the agents. Both a multi-objective optimization technique and a multi-scale continuous loss (MSCL) reward are used to train agents using a progressive penalty to avoid consecutive losses of portfolio value. As a result, better cumulative returns are achieved than when baseline methods are used. In addition, the superiority of our method is emphasized by the result of the bearish test set, where only our method can make a profit. Specifically, our method obtains a 2.36% cumulative return, whereas the baseline methods result in negative cumulative returns. In comparison to FinRL-Ensemble, a reinforcement learning-based method, our method achieves a 46.05% greater cumulative return in the bullish test set.
M. Thamban Nair, Mohamed I. Marie, Laila A. Abd-Elmegid
One of the most significant and extensively utilized cryptocurrencies is Bitcoin (BTC). It is used in many different financial and business activities. Forecasting cryptocurrency prices are crucial for investors and academics in this industry because of the frequent volatility in the price of this currency. However, because of the nonlinearity of the cryptocurrency market, it is challenging to evaluate the unique character of time-series data, which makes it impossible to provide accurate price forecasts. Predicting cryptocurrency prices has been the subject of several research studies utilizing machine learning (ML) and deep learning (DL) based methods. This research suggests five different DL approaches. To forecast the price of the bitcoin cryptocurrency, recurrent neural networks (RNN), long short-term memories (LSTM), gated recurrent units (GRU), bidirectional long short-term memories (Bi-LSTM), and 1D convolutional neural networks (CONV1D) were used. The experimental findings demonstrate that the LSTM outperformed RNN, GRU, Bi-LSTM, and CONV1D in terms of prediction accuracy using measures such as Root Mean Squared Error (RMSE), Mean Absolute Error (MAE), Mean Squared Error (MSE), and R-squared score (R2). With RMSE= 1978.68268, MAE=1537.14424, MSE= 3915185.15068, and R2= 0.94383, it may be considered the best method.
Despite their popularity in recent studies, most hybrid models that exploit the advantages of both classical time series and deep learning models were conducted in univariate forecasting context. For econometric domain which exogenous factors play a crucial role, more studies in multivariate forecasting is essential and should be encouraged. Thus, contributing to hybrid multivariate forecasting literature, a hybrid model named HyBiLSTM was proposed. The algorithm began with ARIMAX GARCHX model forecasting, followed by second forecasting of model residual using Grey Wolf Optimizer based hyperparameters Bidirectional LSTM model. With residuals instead of original multivariate features, LSTM can avoid to processes each feature independently and therefore, reducing convergence complexity and execution time. The final forecasting results was compounded from both models. Three quantitative measurements, Mean Absolute Error (MAE), Root Mean Square Error (RMSE) and Mean Absolute Percentage Error (MAPE), were used to evaluate the established models using the historical daily runoff social and economic based data (01/07/2019-31/12/2022). The findings showed 1) the addition of exogenous factors improved the performance of ARIMA and GARCH models; 2) BiLSTM outperformed other LSTM variants when was integrated with ARIMAX GARCHX model; 3) Using SHAP, Bitcoin price was influenced by stock price, Twitter volume, gold price, and Twitter sentiment index; and 4) structural break had significant effect on forecasting. Other than expanding the literatures regarding hybrid models in multivariate context, this study provides practical contribution for investors by analyzing the factors that the investors can use as an early warning for Bitcoin price fluctuation.
With the expansion of social networks, sentiment analysis has become one of the hot topics in machine learning. However, in traditional sentiment analysis, the text is considered of a general nature and ignores the different aspects that may exist in the text. This paper presents a hybrid model of transfer deep learning methods for the aspect-oriented sentiment analysis of influencers’ tweets to predict the trend of cryptocurrencies. In the first model, different aspects of tweets are extracted using the Concept Latent Dirichlet Allocation (Concept-LDA). Then, by using the pre-trained RoBERTa network and combining it with the Bidirectional Gated Recurrent Unit (BiGRU) deep learning network and attention layer, sentiments of different aspects of tweets are determined. In the following, the price trend of seven cryptocurrencies, Bitcoin, Ethereum, Binance, Ripple, Dogecoin, Cardano, and Solana, is determined using the historical price and the polarity of tweets with BiGRU combined deep neural network and the attention layer. Also, we used the gridsearch method to select dropout hyper-parameters, learning rate, and the number of GRU units, and the Akaike Information Criterion (AIC) criterion confirmed the results of this proposed combination. The results show that the proposed model in the aspect-based sentiment analysis section has been able to achieve 5.94% accuracy and 9.9% improvement in the f1-score on the SemEval 2015 dataset and 2.61% improvement on the SemEval 2016 dataset in f1-score compared to the state-of-arts. Also, the results of predicting the price trend of cryptocurrencies show that the proposed model has correctly recognized the price trend in the next five days in 77% of cases according to the ROC-AUC criterion.
Cryptocurrency as an alternative method of payment that acts both as a type of currency and as a virtual accounting system has always been of interest to investors. Since the public sentiment of a society about cryptocurrencies can affect the cryptocurrencies’ prices, a machine learning model based on sentiment analysis has been proposed to forecast the future prices of cryptocurrencies such as Bitcoin, Ethereum, EOS, Cardano, and Ripple using machine learning models that are suitable for time series data analysis to reduce the risk of investing in this market. It was shown that by applying weights to the sentiment scores of tweets according to the influence factor of the individuals, the accuracy of the prediction will increase and a significant difference between the accuracy scores was observed using the LSTM model according to the MAPE indicator (P=0.045). Also, a hybrid model is proposed based on the combination of features extracted from the texts by one of the dictionary-based text analysis models and the feature of weighted sentiment scores. It was shown that our proposed hybrid model outperformed the other models in predicting the prices of Ethereum, EOS, and Cardano according to the MSE indicator. Also, our proposed model based on weighted sentiment scores according to the influence factor of the Twitterers outperformed the other models in the prediction of the future prices of Bitcoin and Ripple, which indicates that the increase in the number of features will not always lead to an increase in the accuracy of our prediction models.
The purpose of this research is to compare ARIMA and Prophet algorithms and find the best algorithm for forecasting bitcoin prices. The dataset is two years historical bitcoin data between February 2019 and 2021. The data is segmented into daily, weekly, and monthly period category. Both algorithms are built into a univariate model that only receive 2 features for training the model. Several ARIMA models is developed for each dataset interval. After that, the parameter of each model will be cross-referenced to each other to obtain the best parameter combination. Meanwhile, Prophet model will be developed using automatic and manual tuning. Then again parameter value of each model will be cross-referenced to each other to obtain the best parameter combination. Evaluation of the training model is done by calculating the Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), Mean Squared Error (MSE), and Rooted Mean Squared Error (RMSE). The results showed that the best model for the daily and weekly data category was the Prophet algorithm, while for the monthly data category was the ARIMA algorithm.
S. Pourmohammad Azizi, Chien Yi Huang, Ti An Chen, Shu Chuan Chen · 5 authors
<abstract><p>In this article, an alternate method for estimating the volatility parameter of Bitcoin is provided. Specifically, the procedure takes into account historical data. This quality is one of the most critical factors determining the Bitcoin price. The reader will notice an emphasis on historical knowledge throughout the text, with particular attention paid to detail. Following the production of a historical data set for volatility utilizing market data, we will analyze the fundamental and computed values of Bitcoin derivatives (futures), followed by implementing an inverse problem modeling method to obtain a second-order differential equation model for volatility. Because of this, we can accomplish what we set out to do. As a direct result, we will be able to achieve our objective. Following this, the differential equation of the second order will be solved by an artificial neural network that considers the dataset. In conclusion, the results achieved through the utilization of the Python software are given and contrasted with a variety of other research approaches. In addition, this method is determined with alternative ways, and the outcomes of those comparisons are shown.</p></abstract>