Holistic and proactive machine learning approaches to predict cyberattacks in blockchain networks

This paper presents a holistic and proactive machine learning framework for predicting cyberattacks in blockchain networks. The study adopts the Hawkes process as a self-exciting point process model to generate realistic attack event sequences that capture clustering and contagion effects. By integrating the blockchain governance game with machine learning techniques, an innovative defense architecture is proposed. A ResNet-based attack predictor is developed to accurately estimate the optimal timing for releasing reserved honest nodes before an attacker exceeds the 51 percent threshold. A strategic decision engine maps these predictions to three operational modes, enabling intelligent real-time resource allocation and dynamic network scale adjustment. The framework effectively reduces security risks while maintaining operational continuity. All source codes and datasets are publicly available to support reproducibility and future research in machine learning-enabled blockchain security.