SECURITY-ORIENTED FORECASTING OF FINANCIAL TRENDS IN THE CRYPTOCURRENCY MARKET USING MACHINE LEARNING MODELS

Abstract

The rapid expansion of the cryptocurrency market has introduced new opportunities for investors and digital economy stakeholders while simultaneously amplifying systemic risks due to extreme price volatility and decentralized regulation. This study addresses the challenge of security-oriented forecasting of cryptocurrency price trends by applying advanced machine learning techniques to support decision-making under uncertainty. The research aims to design and evaluate predictive models capable of providing accurate short-term forecasts, thereby mitigating potential financial losses in volatile market conditions. To achieve this, we implement and compare three forecasting approaches: Ridge regression, Lasso regression, and Long Short-Term Memory (LSTM) neural networks. Historical data for Bitcoin (BTC), Cardano (ADA), and Tether (USDT) from 2020 to 2024 were used to test the models, representing high, medium-, and low-volatility assets, respectively. The results reveal that the LSTM model significantly outperforms linear models in scenarios characterized by sharp price fluctuations, particularly for Bitcoin and Cardano. After hyperparameter tuning via grid search, LSTM demonstrated the lowest Mean Absolute Error (MAE) and Root Mean Square Error (RMSE), confirming its superior ability to capture complex temporal dependencies and nonlinear patterns. Conversely, Ridge and Lasso regression exhibited robust performance for stable assets such as Tether, ensuring reliable predictions with reduced computational costs. The findings highlight the necessity of an adaptive, asset-specific modeling strategy for effective risk management in financial forecasting. Furthermore, the integration of machine learning-based predictive models into automated monitoring and alert systems could enhance the resilience of socioeconomic systems against market instability. This research contributes to the development of security-oriented decision-support frameworks by providing empirical evidence and methodological guidelines for applying machine learning in high-risk financial environments.