This paper investigates the role of artificial intelligence (AI) stocks and AI cryptocurrencies in portfolio diversification, reflecting on the rising interest in technology-oriented assets. While much research has focused on the diversification, hedging, and safe-haven properties of digital assets, such as Bitcoin and Ethereum, this study focuses on whether AI cryptocurrencies and AI stocks provide untapped diversification potential. Using mean-variance, risk parity, and higher-order moments approaches, we construct portfolios that combine AI stocks, AI cryptocurrencies, and traditional assets under various optimization frameworks. The findings reveal that the mean-variance framework is more conservative in allocating to AI cryptocurrencies, while the higher-order moments approach accommodates for greater flexibility. Seemingly, investors may benefit from expanding their asset pool to incorporate AI stocks and AI cryptocurrencies. Across most portfolio settings, gold and commodities dominate allocations, followed by AI stocks, with AI cryptocurrencies receiving only marginal weights owing to their high volatility. However, allocations to AI cryptocurrencies increase as investor risk tolerance increases, thereby highlighting their potential for risk-seeking portfolios. Overall, the results indicate that AI stocks and AI cryptocurrencies can enhance portfolio diversification and improve risk-return outcomes. These results offer valuable insights for investors seeking to optimize their portfolios, through exposure to emerging technology-driven assets while balancing traditional risk considerations. • The study explores the diversification potential of AI Stocks and AI Cryptocurrencies to a traditional portfolio. • Dominated by NVIDIA and Tesla, AI stocks perform better than AI cryptocurrencies. • AI cryptocurrencies offer limited diversification benefits while significantly increasing portfolio risk. • Unlike AI stocks, AI cryptocurrencies are not dominated by a single player in portfolio diversification. • Allocation to AI cryptocurrencies is highly sensitive to investor risk aversion, particularly driven by their explosive market behaviour.
The rise of Decentralized Finance (DeFi), enabled by blockchain technology, has introduced open and transparent financial ecosystems that contrast sharply with traditional financial systems. While DeFi expands the financial landscape and democratizes participation in global markets, it also introduces new complexities. Classic valuation models used in traditional finance often fall short in this context. However, DeFis transparency---where all transactions are publicly recorded---offers a unique opportunity to model and understand market behavior using modern analytical tools. Motivated by the challenges and opportunities of DeFi, the main goal of this thesis is to propose novel methods to understand market dynamics through the lens of network science and machine learning. To this end, we focus on four specific objectives: (i) assess whether structural information from blockchain transaction networks provides predictive signals beyond traditional indicators; (ii) develop robust trust-based valuation metrics for DeFi protocols; (iii) develop a framework for forecasting financial time series through uncertainty-aware machine learning architectures; (iv) construct diversified portfolios using network-based representations of asset relationships. First, using Ethereum as a case study, we analyze the influence of the transaction network on market trends by comparing the performance of two machine learning models: one that uses technical analysis and social media indicators commonly found in the literature and another that incorporates structural properties of the transaction network. We found that by including transaction network information, we can anticipate 46% more uptrends and 19% more downtrends, highlighting the predictive power of the transaction network. Second, we introduce the TVL/MCAP bands as a tool to identify periods of overconfidence and underconfidence in the DeFi market. We show that extreme values of this indicator can signal price movements: values above the 95th percentile are associated with a 15\% higher return in the following month, while values below the 5th percentile anticipate declines, highlighting investor confidence as a key market driver. Third, we address the need for forecasts that not only anticipate market trends but also quantify the uncertainty surrounding them. To this end, we integrate Reservoir Computing (RC) with conformal prediction methods to provide statistically rigorous forecasts along with prediction intervals. We found that RC outperform traditional econometric models, particularly in anticipating the trend of financial time series. Furthermore, we show that conformal methods, especially quantile-conformal variants, significantly improve forecast reliability while adapting to market volatility. Finally, we address the challenge of portfolio optimization using network-based methods. Specifically, we model the network of relationships between cryptocurrencies to obtain a market representation that enables selecting a more diversified portfolio. We find that peripheral assets enhance portfolio stability and returns, while links bridging network communities carry higher risk. Thereby, these results highlight the importance of structural diversification in volatile markets. In addition, we contribute to refining pairs trading strategies by proposing the Hurst exponent to identify rapid mean-reversion opportunities. We show that anti-persistent values of H lead to faster reversion and consistent returns---minimizing trading costs and enabling broader portfolio construction. In conclusion, this thesis provides an interdisciplinary analytical framework that advances our understanding of DeFi markets. By introducing network-based indicators, trust metrics, uncertainty-aware forecasts, and diversification strategies grounded in market structure, we provide new tools for investors and researchers to navigate the complexity and volatility inherent in decentralized financial systems. RESUMEN El auge de las Finanzas Descentralizadas (DeFi), impulsado por la tecnología blockchain, ha dado lugar a ecosistemas financieros más accesibles y transparentes que contrastan con los sistemas financieros tradicionales. DeFi amplía el panorama financiero actual e introduce nuevos retos, como la necesidad de un nuevo modelo de valoración de los activos. No obstante, el hecho de que todas las transacciones son públicas, ofrece una oportunidad única para modelar y comprender la dinámica del mercado mediante nuevas herramientas analíticas. Esta tesis tiene como objetivo principal proponer nuevos métodos para comprender la dinámica del mercado desde la perspectiva de los sistemas complejos y el aprendizaje automático. Para ello, nos centramos en cuatro objetivos específicos: (i) evaluar si la información estructural de las redes de transacciones aporta señales predictivas más allá de los indicadores tradicionales; (ii) desarrollar métricas de valoración de los protocolos DeFi basadas en la confianza de los inversores; (iii) construir un marco metodológico para predecir series temporales financieras mediante arquitecturas de aprendizaje automático que incorporen incertidumbre; (iv) construir portfolios diversificados utilizando representaciones de la red de relaciones entre criptomonedas. En primer lugar, utilizando Ethereum como caso de estudio, analizamos la influencia de la red de transacciones sobre la tendencia del mercado comparando dos modelos de aprendizaje automático: uno que emplea indicadores de análisis técnico y de redes sociales comunes en la literatura, y otro incluyendo propiedades estructurales de la red de transacciones. Los resultados muestran que incluyendo información de la red podemos anticipar un 46% más de tendencias alcistas y un 19% más de tendencias bajistas, lo que subraya el poder predictivo de la red de transacciones. En segundo lugar, introducimos las bandas TVL/MCAP para identificar períodos de sobreconfianza y desconfianza en el mercado DeFi. Demostramos que valores extremos de este indicador anticipan movimientos en el precio: valores por encima del percentil 95 se asocian con un rendimiento 15% superior en el mes siguiente, mientras que valores por debajo del percentil 5 anticipan caídas. En tercer lugar, abordamos la necesidad de predicciones que no solo anticipen tendencias del mercado, sino que también cuantifiquen la incertidumbre. Para ello, integramos Reservoir Computing (RC) con métodos de predicción conforme para generar predicciones estadísticamente rigurosas junto con intervalos de confianza. Mostramos que RC supera a los modelos econométricos tradicionales, especialmente anticipando la tendencia del precio. Además, los métodos conformes ---en particular las variantes de cuantiles--- mejoran significativamente la fiabilidad de las predicciones al adaptarse a la volatilidad del mercado. Por último, abordamos el problema de optimización de portfolios mediante métodos basados en redes. Específicamente, modelamos la red de relaciones entre criptomonedas para seleccionar un portfolio más diversificado. Observamos que evitar pares que conectan distintas comunidades en la red y priorizar activos periféricos aumenta el rendimiento y disminuye el riesgo, demostrando así la importancia de una diversificación estructural. Además, proponemos el uso del exponente de Hurst para identificar oportunidades que revierten antes a la media en estrategias de pairs trading. En conclusión, esta tesis propone un marco analítico interdisciplinar que contribuye al entendimiento de los mercados DeFi. Al introducir indicadores basados en redes, métricas de confianza, predicciones con estimación de incertidumbre y estrategias de diversificación basadas en la estructura del mercado, ofrecemos nuevas herramientas para que inversores e investigadores naveguen la complejidad y volatilidad propias de los sistemas financieros descentralizados.
ABSTRACT This paper employs deep learning and machine learning‐based NLP models to investigate the impact of the news sentiment on the Bitcoin price. The lagged Bitcoin variables, news indicators, macroeconomic, and financial factors were taken into account to explain the importance of news sentiment on the Bitcoin price. Moreover, FinBERT‐based sentiment scores and semantic features extracted from over 650,000 financial news headlines were integrated with financial and macroeconomic variables. The importance scores of the investigation showed that Bitcoin was largely explained by its lagged price movements, which suggests the speculative nature of the cryptocurrency. However, the investigation also revealed that Bitcoin was significantly influenced by the news sentiment score. In other words, the paper indicates that the movements in the Bitcoin price can be predominantly explained by the news sentiment. Advanced hybrid models (all ML and DL models with the addition of variables obtained with the FinBERT model) were optimized using Optuna and RandomizedSearchCV. The FinBERT‐LSTM model achieved the best prediction accuracy. Nevertheless, the main findings indicated that the response of the Bitcoin price to negative news was much stronger than to positive and neutral news. This finding suggests that the asymmetric relationship between the Bitcoin price and news sentiment was evident. GARCH‐based volatility and what‐if scenario analyses further demonstrated that negative sentiment leads to sharper fluctuations in the Bitcoin price. The paper provides important implications for policymakers, portfolio managers, investors, and academics.
This study investigates the application of the Light Gradient Boosting Machine (LGBM) model for both deterministic and probabilistic forecasting of Bitcoin realized volatility. Utilizing a comprehensive set of 69 predictors -- encompassing market, behavioral, and macroeconomic indicators -- we evaluate the performance of LGBM-based models and compare them with both econometric and machine learning baselines. For probabilistic forecasting, we explore two quantile-based approaches: direct quantile regression using the pinball loss function, and a residual simulation method that transforms point forecasts into predictive distributions. To identify the main drivers of volatility, we employ gain-based and permutation feature importance techniques, consistently highlighting the significance of trading volume, lagged volatility measures, investor attention, and market capitalization. The results demonstrate that LGBM models effectively capture the nonlinear and high-variance characteristics of cryptocurrency markets while providing interpretable insights into the underlying volatility dynamics.
Ala Alrawajfi, Mohd Tahir Ismail, Sadam Al Wadi, Saleh Atiewi
Ethereum and other cryptocurrencies are volatile, making Ethereum-USD rate evaluation difficult.Due to unsuccessful data collection and exchange downtimes, financial time series data are incomplete and lacking critical values.Thus, assessments may be incomplete, and trends may be miscalculated.This research builds and tests an ARIMA-random forest data imputation method to overcome these concerns.This innovative strategy uses AutoRegressive Integrated Moving Average (ARIMA) to describe the linear chronologic sequence relationship and random forest to solve nonlinearity.The suggested method uses ARIMA to handle the linear time-dependent data feature and random forest to reduce estimation errors to improve Ethereum-USD closing price estimates.The mean absolute error (MAE) and mean absolute percentage error (MAPE) results demonstrate that the proposed hybrid model significantly outperforms conventional imputation approaches across all missing data levels (10%-50%).For example, at 30% missing data, the hybrid model achieved an MAE of 0.91 and a MAPE of 0.00074, compared to ARIMA's MAE of 2.21 (MAPE 0.00185) and Random Forest's MAE of 2.34 (MAPE 0.00186).Across all scenarios, the hybrid model reduced MAE by up to 60% and MAPE by over 55% relative to the best single-method baseline, indicating superior robustness and accuracy in handling incomplete Ethereum-USD datasets.By providing precise market and result knowledge, these insights help financial analysts, traders, and researchers make accurate, efficient decisions.
The rapid evolution of financial technologies (FinTech) and digital assets—including cryptocurrencies, decentralized finance (DeFi), and tokenized capital markets—has created an unprecedented need for secure, scalable, and computationally efficient systems. This study examines the transformative potential of quantum computing in reshaping financial technology infrastructures and digital asset ecosystems. Traditional computational models, constrained by classical encryption limits and the exponential growth of financial data, face increasing inefficiencies in handling real-time risk assessment, portfolio optimization, and transaction verification. Quantum computing, with its capacity for superposition, entanglement, and parallel state evaluation, provides novel opportunities to redefine data security, financial modeling, and cryptographic mechanisms in the digital economy. The research explores how quantum algorithms—notably Quantum Approximate Optimization Algorithm (QAOA), Quantum Fourier Transform (QFT), and Grover’s search algorithm—can enhance financial operations such as market forecasting, fraud detection, and asset pricing. Additionally, it investigates quantum-resistant cryptography to safeguard digital asset networks against the vulnerabilities introduced by future quantum decryption capabilities. By integrating hybrid quantum–classical frameworks, this approach enables the development of sustainable, adaptive, and transparent financial systems. The findings highlight quantum computing’s potential to advance financial inclusion, increase transaction speed, and improve systemic resilience. As global financial markets transition toward quantum readiness, the convergence of FinTech and quantum innovation is expected to redefine how digital assets are managed, traded, and secured—marking a paradigm shift toward quantum financial intelligence.
This paper presents the revolutionary Kera Protocol, a mathematically proven blockchain architecture that fundamentally solves the dual crises of accessibility and sustainability plaguing contemporary decentralized finance systems. The work addresses the stark reality that 99.95% of humanity remains excluded from blockchain validation due to prohibitive capital requirements. The paper's core contribution establishes the first provably sustainable economic model in blockchain history through a dynamic APY allocation algorithm that maintains the fundamental invariant OBLIGATIONS = REVENUE at every 12-second block interval. This mathematical constraint creates theoretical impossibility of protocol insolvency, directly addressing the $108 billion in losses from failed DeFi protocols like Terra/LUNA, Celsius, and BlockFi that promised unsustainable fixed returns. The research introduces an innovative vault-to-pool economic architecture leveraging 20x capital efficiency to deliver mathematically certain 102% APY returns—derived from real interest accrual rather than speculative mechanisms. Rigorous validation through the MALIV (Multi-Agent Long-term Investment Validator) model simulates 14,600 days across 40 years, incorporating realistic market cycles, black swan events (0.5% probability), and extreme stress scenarios including 99% revenue drops. Across 3,000+ simulation runs, the protocol demonstrated 100% sustainability with perfect equality maintenance. The paper details seven diversified revenue streams projected to scale from $87 million in Year 1 to $35.75 billion by Year 5, eliminating reliance on inflationary tokenomics. Technical innovations include autonomous validator bot systems that eliminate slashing risks, browser-based validation infrastructure, and deflationary token buyback mechanisms. The work represents PhD-level contributions to solving the DeFi Sustainability Trilemma, with planned submissions to leading academic journals in financial economics and computational economics.
We investigate a multi-class machine learning (ML) framework to generate daily Bitcoin trading signals—Buy, Sell, or Hold. Three algorithms—XGBoost, LightGBM, and Random Forest—are compared with a naive buy-and-hold strategy. Using BTC/USD daily data (2015–2024), we apply a range of technical indicators across trend, momentum, volatility, and volume, later pruned by correlation analysis. A ±1% threshold defines the "Hold" zone to avoid minor fluctuations. Empirical tests show that LightGBM outperforms other models and even surpasses buy-and-hold in final portfolio value. Our findings support the design of tri-class ML strategies tailored for high-volatility markets like cryptocurrency.
This study examines the intricate relationships between cryptocurrency and various uncertainties related to economic policy and global risk factors. It explores the interactions between cryptocurrency and global risk factors, comparing these with their relationships to different measures of economic policy uncertainty (EPU). We find that cryptocurrency returns are more sensitive to global risk factors than to the country-level EPU. Notably, gold exhibits bidirectional causality with cryptocurrency in returns and volatility. The research sheds light on the dynamic interactions within cryptocurrency markets, underscoring the importance of continuous monitoring and adaptive strategies to navigate the evolving financial landscape of the digital ecosystem.
This study explores how integrating cryptocurrencies into traditional financial portfolios can influence investment performance. Focusing on Bitcoin and Ethereum alongside key European stock indices (BUX, DAX, and FTSE), the analysis examines whether blockchain-based assets can enhance diversification and improve the balance between risk and return. Using weekly market data from 2019 to 2023, the research applies Markowitz mean–variance optimization to identify optimal asset allocations under different objectives such as maximizing the Sharpe ratio, minimizing risk, and maximizing returns. The findings reveal that cryptocurrencies show weak correlations with European stock indices, suggesting meaningful diversification potential. When included in portfolios, Bitcoin and Ethereum can significantly boost returns, though they also increase volatility. Portfolios optimized for risk reduction favored traditional indices, while those targeting higher returns relied predominantly on cryptocurrencies. Overall, combining digital and conventional assets produced a more balanced performance, with the Sharpe-ratio–maximized portfolio demonstrating the best trade‐off between stability and profitability. These results indicate that cryptocurrencies can play a valuable complementary role in modern portfolio construction. They are most suitable for investors willing to accept higher risk in exchange for potentially greater rewards, while more risk‐averse investors may benefit from maintaining a stronger focus on traditional equity indices. The study contributes to understanding how blockchain‐driven assets can expand financial opportunities and supports a broader view of diversification in contemporary investment strategies.
This paper evaluates the performance of classical time series models in forecasting Bitcoin prices, focusing on ARIMA, SARIMA, GARCH, and EGARCH. Daily price data from 2010 to 2020 were analyzed, with models trained on the first 90 percent and tested on the final 10 percent. Forecast accuracy was assessed using MAE, RMSE, AIC, and BIC. The results show that ARIMA provided the strongest forecasts for short-run log-price dynamics, while EGARCH offered the best fit for volatility by capturing asymmetry in responses to shocks. These findings suggest that despite Bitcoin's extreme volatility, classical time series models remain valuable for short-run forecasting. The study contributes to understanding cryptocurrency predictability and sets the stage for future work integrating machine learning and macroeconomic variables.
This research examines deep-learning and machine-learning models for cryptocurrency price prediction, with a keen focus on Bitcoin (BTC), Ethereum (ETH), Ripple (XRP), and Solana (SOL). Cryptocurrencies exhibit high volatility, non-linear behavior and are able to react strongly to exogenous events, making their prediction and forecasting challenging. The primary aim of this research is to determine which predictive models yield optimal performance in characterizing these complexities and to provide empirical guidance on real-life investment and risk-management applications. Four approaches were used for this forecasting: Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), a combination of LSTM-GRU models, and Stochastic Gradient Descent (SGD) regression. The daily historical data were used to train and test each model on different forecast horizons, and performance was measured accordingly by Mean Squared Error (MSE) and Mean Absolute Error (MAE) values. As shown in the results, it can be observed that GRU exhibited the lowest error rates in the majority of the assets, particularly in short-term predictions. LSTM demonstrated a promising ability to capture long dependencies, whereas the hybrid LSTM-GRU system showed a similar performance proficiency by combining the relative superiorities of the two respective models. On the other hand, the conventional SGD regression was the worst among all the deep-learning algorithms, thereby demonstrating the extreme capability of these algorithms in modelling non-linear time sequences. The results confirm GRU as the most viable model for AI-powered crypto prediction and demonstrate the potential of hybrid architecture, at least in certain situations. This study will contribute to the existing debates about the role of deep learning in predicting financial outcomes and provide valuable insights to traders, analysts, and researchers navigating the uncertainties of the digital asset world.
Cryptocurrency markets are characterized by high volatility and complex patterns, creating both challenges and opportunities for traders and investors. This study introduces a machine learning framework for cryptocurrency trading optimization that leverages advanced analytical techniques to enhance trading decisions. We extracted historical data for 30 cryptocurrencies over a four-year period from Yahoo Finance. After preprocessing, we applied Principal Component Analysis (PCA) and K-means clustering to select representative coins. Four machine learning models (Gradient Boosting, XGBoost, Support Vector Regression, and Long Short-Term Memory networks) were trained to predict cryptocurrency price movements. Model performance was evaluated using multiple metrics, including Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R-squared (R 2 ). Gradient Boosting and XGBoost consistently outperformed SVR and LSTM models across all cryptocurrencies, with R 2 values of approximately 0.98 for most coins. The framework successfully identified trading signals through both moving average strategies and machine learning predictions, providing actionable insights for cryptocurrency traders. Our analysis demonstrates that ensemble-based models offer superior performance for cryptocurrency price prediction compared to neural network approaches. The integration of advanced visualization tools and trading signal generation creates a comprehensive system for data-driven cryptocurrency trading decisions.
Abstract This research paper provides a comprehensive analysis of Bitcoin, the world’s preeminent cryptocurrency, focusing on the economic drivers of its price formation, its broader impact on the economy, and the evolving dynamics of its volatility. Drawing on high-frequency econometric modeling, time-series analysis, and network-based prediction methods, the paper synthesizes insights from leading empirical studies to elucidate the factors shaping Bitcoin’s price, including supply-demand fundamentals, investor behavior, macro-financial indicators, transaction network structure, and the influence of derivative markets. Additionally, it explores Bitcoin’s adoption in key industries, its intrinsic and extrinsic value determinants, and the implications of its volatility for financial stability. The study concludes by reflecting on the future trajectory of Bitcoin as it transitions from speculative asset to potential mainstream medium of exchange, considering regulatory, technological, and market challenges. Keywords: Bitcoin, cryptocurrency, price formation, volatility, supply-demand, GARCH, partial differential equations, transaction networks, futures markets, economic impact
This paper investigates the relationship between cryptocurrencies and other financial assets, with a particular focus on the dynamics of information flow between developed and emerging markets. To achieve this objective, the study applies a combined methodology of spillover index analysis and network topology based on graph theory. The analysis covers key cryptocurrencies (Bitcoin and Ethereum), stocks, and conventional currencies over the period November 2017 to September 2022, and distinguishes between short-term and long-run dynamics. The empirical findings show that in the short run, Bitcoin and Ethereum predominantly act as net shock transmitters, whereas in the long run, stocks and conventional currencies, together with Bitcoin and Ethereum, become the principal conveyors of spillover shocks. The network topology analysis corroborates these results by revealing the centrality of these assets in the spillover structure. By integrating spillover and network approaches across different markets and time horizons, this study contributes to the literature by providing a more nuanced understanding of how cryptocurrencies interact with traditional financial assets under varying market conditions.
This research paper presents the design and development of a Cryptocurrency Dashboard that applies data analytics techniques to the financial technology sector. The goal of this project is to visualize historical cryptocurrency data such as market capitalization, trading volume, and price fluctuations through an interactive and user-friendly interface. Using tools such as Python and Power BI, data was collected, cleaned, analyzed, and visualized to provide dynamic insights for investors and analysts. The dashboard enables efficient decision-making by simplifying complex financial data into clear and interpretable visuals. The study demonstrates how data analytics enhances understanding of cryptocurrency trends and contributes to evidence-based financial analysis in the digital economy.
Sr IT Developer, First Horizon Bank, Memphis, TN, USA, Rushikesh Anantrao Deshpande
The article examines methods for optimizing PL/SQL queries in distributed banking databases, emphasizing the transition from static rule-based mechanisms to adaptive, learning-driven architectures. The study’s relevance is defined by the increasing complexity of financial data environments that require real-time consistency, fault tolerance, and intelligent workload distribution. The research synthesizes results from seven recent works published between 2021 and 2025, covering neural cost modeling, heuristic algorithms, hybrid plan enumeration, and visualization-based diagnostics. Special attention is devoted to learned cost models and metaheuristic strategies that enhance selectivity estimation, reduce latency, and stabilize throughput in distributed ledger systems. The methodological framework integrates comparative analysis, systematization, and critical evaluation of hybrid, heuristic, and learning-based optimizers. The findings reveal a multi-layered optimization model that combines probabilistic inference, robust plan selection, and heuristic refinement. The conclusions underscore the practical applicability of adaptive PL/SQL optimization for high-volume banking infrastructures and data-intensive financial analytics.
Rabia Arshad, Muhammad Milhan Afzal Khan, Saman Rasheed, Irtaza Ijaz · 5 authors
Blockchain technology has transformed decentralized data exchange and digital payments but the consistently high gas prices pose a significant challenge to its scalability and efficiency. This research explores the role of AI-driven gas price prediction and data compression methods on gas utilization in blockchain systems with special emphasis on Ethereum transactions. Using actual Ethereum transaction history, we compare the performance of compressed versus uncompressed payloads with three different compression algorithms: Zlib, Brotli, and Gzip. Beyond that, a linear regression model is also trained to forecast hourly gas Price fluctuations given past transaction history. The methodology includes thorough statistical analysis to provide accurate and reproducible results. Our results show that compressing text data over 141 bytes using the Zlib algorithm prior to making transactions on the Ethereum network decreases the amount of gas Used without altering system time. This validates the efficiency of combining data compression with gas price forecasting in minimizing transaction costs without affecting performance. Moreover, our study further encompasses investigation of actual gas Price trends and provides real-world insights for optimizing timing strategies for economic transaction execution. These results enhance the knowledge of Ethereum gas dynamics and provide valuable solutions for enhancing economic efficiency and resource utilization in applications based on blockchain. Future efforts will involve applying the framework to the Ethereum mainnet, using deep learning models for increased prediction accuracy, and adaptive compression dependent on network state and transaction size.
Cryptocurrency price prediction poses significant challenges due to the inherent volatility and nonlineardynamics of the market. This study introduces a hybrid stacked modeling framework that integrates machine learning (ML) and deep learning (DL) techniques, capitalizing on their complementary strengths-ML models are effective at capturing nonlinearfeature interactions in structured data, while DL architectures are adept at modeling temporal dependencies in sequential data. The proposed model leverages historical price data, technical indicators, macroeconomic variables, and sentiment metrics, with feature engineering applied to enhance predictive capability. Empirical evaluation was conducted through two experimental setups: (i) short-term, monthly segment analysis and (ii) long-term generalization via five-fold cross-validation. The hybrid model outperformed individual baseline models, achieving up to 18.3% lower RMSE and 6.7% higher directional accuracy. Additionally, it yielded superior risk-adjusted returns, with Sharpe Ratios reaching 0.094 on the Ethereum dataset. Beyond technical improvements, this research offers foresight into digital financial markets, providing a robust tool for investors, institutions, and policymakers navigating the evolving cryptocurrency landscape. The model supports more informed decision-making, enhances market oversight, and contributes to the development of adaptive regulatory frameworks for digital finance.
Alexandr Kuznetsov, Олексій Костенко, K.O. Klymenko, Zoriana Hbur · 5 authors
Blockchain-based cryptocurrency markets present unique analytical challenges due to their decentralized nature, continuous operation, and extreme volatility. Traditional price prediction models often struggle with the binary trade execution problem in these markets. This study introduces a confidence-based classification framework that separates directional prediction from execution decisions in cryptocurrency trading. We develop a neural network system that processes multi-scale market data, combining daily macroeconomic indicators with a high-frequency order book microstructure. The model trains exclusively on directional movements (up versus down) and uses prediction confidence levels to determine trade execution. We evaluate the framework across 11 major cryptocurrency pairs over 12 months. Experimental results demonstrate 82.68% direction accuracy on executed trades with 151.11-basis point average net profit per trade at 11.99% market coverage. Order book features dominate predictive importance (81.3% of selected features), validating the critical role of blockchain microstructure data for short-term price prediction. The confidence-based execution strategy achieves superior risk-adjusted returns compared to traditional classification approaches while providing natural risk management capabilities through selective trade execution. These findings contribute to blockchain technology applications in financial markets by demonstrating how a decentralized market microstructure can be leveraged for systematic trading strategies. The methodology offers practical implementation guidelines for cryptocurrency algorithmic trading while advancing the understanding of machine learning applications in blockchain-based financial systems.
Modern economic ecosystems require radical hazard management systems that may take care of big streams of statistics without compromising on regulatory compliance and business transparency. Conventional batch-based risk assessment models exhibit intrinsic shortcomings in addressing millisecond-level market turbulence and intricate network interdependencies that define new trading environments. Sophisticated artificial intelligence platforms embedded in distributed computing environments offer transformational possibilities for real-time risk sensing and mitigation. The suggested architecture develops end-to-end risk analytics capacity via ensemble machine learning algorithms, graph contagion analysis, and explainable AI features to meet strict regulatory demands. Complex data pipelines ingest heterogeneous finance streams from worldwide exchanges, payment networks, and blockchain ledgers in tandem. Tailored graph neural networks examine systemic risk transmission patterns in connected financial institutions while retaining dynamic relationship mapping capabilities. Explainable AI integration presents version interpretability and regulatory adherence through function attribution strategies and robust audit trail retention. Cloud-local infrastructure layout helps elastic scaling throughout multi-cloud environments using fault-tolerant distributed orchestration systems. Performance assessments display large upgrades in detection latency and predictive accuracy relative to standard batch-processing strategies. The design embodies a paradigm shift towards forward-looking, adaptive, and transparent risk management functionality critical to ensuring financial stability in progressively complex market conditions
El estudio analizó el comportamiento de las criptomonedas Bitcoin y Ethereum durante el año 2024 mediante la construcción de un modelo estadístico ARIMA (AutoRegressive Integrated Moving Average). La investigación utilizó un enfoque cuantitativo que se dividió en cuatro etapas: recopilación y limpieza de datos históricos, verificación de la estabilidad de los datos, identificación y estimación de los mejores parámetros usando criterios de información, y validación mediante medidas de precisión y análisis de residuos. Los resultados demostraron que el modelo ARIMA fue útil en el pronóstico de valores en mercados estables, destacando su trayectoria en el análisis de datos financieros. Además, los valores bajos de RMSE y MAPE validaron que el modelo tiene la capacidad de realizar pronósticos precisos en escenarios con alta frecuencia. En particular, el MAPE de Bitcoin fue 2,25 % y el de Ethereum 2,85 % durante la etapa de prueba, demostrando que los valores pronosticados tuvieron una ligera desviación con respecto a los reales. No obstante, el modelo puede verse afectado en periodos de alta volatilidad, como en las burbujas especulativas o en los desplomes bursátiles, ya que no tiene la capacidad de adaptarse dinámicamente a cambios súbitos en los parámetros; sin embargo, su utilidad puede mejorar al combinar modelos híbridos con ARIMA.
Samsudin Samsudin, Muhammad Dedi Irawan, Muhammad Irwan Padli Nasution, Raissa Amanda Putri
Bitcoin’s extreme price volatility has long posed challenges for both investors and researchers seeking reliable forecasting models. Conventional financial approaches often fail to capture the highly complex, nonlinear, and fast-moving nature of cryptocurrency markets. To address this gap, this study develops a Bitcoin price prediction model using Random Forest Regression based on on-chain market data. The dataset was obtained from publicly available historical Bitcoin daily trading records spanning more than five years. Key features include opening price, daily high and low ranges, trading volume, and percentage change. The research was carried out in several stages. First, data preprocessing was conducted through normalization, handling of missing values, and feature engineering. Second, model training was performed with Random Forest, including parameter tuning to optimize predictive accuracy. Third, model evaluation employed R² and Mean Absolute Percentage Error (MAPE) as primary performance indicators. Fourth, visualization was implemented using interactive charts to allow users to observe short-term price fluctuations and long-term market patterns. The system development followed an iterative methodology inspired by the Streamlit Framework, which is an open-source Python library that simplifies building interactive web applications for data science and machine learning. This approach provides flexibility, enabling rapid experimentation and adaptation to evolving market conditions. The results show that the proposed model achieves near-perfect R² values (approaching 1.0) with consistently low MAPE, highlighting its reliability. Beyond predictive performance, the framework is designed to be scalable, supporting future integration with deep learning methods such as LSTM and external macroeconomic indicators, thus offering both practical utility for investors and academic contributions to decentralized finance research.
Puguh Hiskiawan, Jovan William, Louis Feliepe Tio Jansel
Bitcoin, a highly volatile and decentralized digital asset, presents considerable challenges for accurate price forecasting. This study proposes an applied data science framework that compares traditional statistical approaches with modern Artificial Intelligence (AI)-based models to predict Bitcoin’s daily closing price. Using BTC-USD historical data from January 2020 to December 2024, we converted prices into Indonesian Rupiah (IDR) to increase local relevance. Our forecasting horizon is 30 days, based on a 60-day lookback window. We evaluate six models: Linear Regression, ARIMA, and Prophet as traditional techniques, alongside Random Forest, XGBoost, and Long Short-Term Memory (LSTM) networks as AI approaches. All models were trained using lag-based or sequence-based time series features and evaluated using MAE, RMSE, R², MAPE, and SMAPE. Results show that AI models, particularly LSTM and XGBoost, offer better performance in capturing short-term non-linear dynamics compared to traditional models. LSTM provides high accuracy, though with greater computational demand, while XGBoost strikes a balance between speed and precision. Prophet and ARIMA remain effective for quick and interpretable forecasts but struggle with abrupt trend shift common in cryptocurrency markets. In addition to performance metrics, we include a robustness analysis based on median absolute error and outlier detection to assess model stability under extreme variations. Visual analytics—including forecast curves, error distributions, and uncertainty bounds—help interpret and communicate model behavior. This comprehensive evaluation offers practical insights for investors, analysts, and fintech practitioners, and the pipeline can be extended to other volatile assets.