Evaluating Port Emissions Prediction Model Resilience Against Cyberthreats

The port sector is evolving, influencing areas such as operations, infrastructure, governance, technology, partnerships, and commercial strategies. AI can significantly improve the efficiency of intermodal freight transportation by enhancing real-time operational processes and decision-making models. This research evaluates the resilience of machine learning models to various data and architecture threats, focusing on adversarial attacks and data corruption. The study explores a range of attack methods, including white-box attacks, data poisoning, and model evasion, and tests defensive strategies like adversarial training, resilient optimization, and anomaly detection on a developed DNN model. The findings highlight the vulnerabilities of current models and emphasize the need for robust security measures. The paper concludes with recommendations for enhancing model robustness and suggests directions for future research to bolster AI systems against evolving threats, ensuring their reliability and security in unpredictable environments.