Enhancing Marine Radar Security Through Semi-Supervised Learning: A Self-Training Approach

Marine radar systems are essential for maritime safety but are increasingly vulnerable to cyber-attacks. This research presents a method to improve marine radar security using semi-supervised learning with a Self-Training Classifier. By leveraging both labeled and unlabeled data, this approach enhances the detection of cyber threats. The RadarPWN dataset, consisting of network traffic from marine radar systems, was preprocessed and used to train a RandomForestClassifier (RFC). Through iterative training, high-confidence predictions from unlabeled data were incorporated to improve detection accuracy. The method achieved 99.99% accuracy, effectively identifying multiple attack types, including scaling, addition, rotation, move, translation, removal, and Denial of Service (DoS) attacks. This approach strengthens the resilience of marine radar systems, addressing the challenge of limited labeled data in maritime cybersecurity.