Intelligent Cyber-Physical Protection for Maritime IoT Ecosystems: Machine Learning-Based Intrusion Detection for Intelligent Marine Systems

Abstract

In the modern network security study, the intrusion detection systems (IDS) play the central role in securing the Internet of Things (IoT) infrastructures against cyber-attacks like the Distributed Denial of Service (DDoS) and the brute force attacks. But there are often difficulties of imbalance classes and high-dimensional data faced by traditional IDS.This study proposes a real-time, machine learning-based framework tailored for the maritime edge, delivering up to 99.8% detection accuracy while ensuring compliance with international cyber risk management guidelines. As part of on-going research, we attempted to come up with good intrusion detection models involving machine learning (ML). Having chosen the UNSW-NB 15 and CIC IoT 2023 benchmark datasets, all of which reflect the cyber threats observed in the real world, we tested the performance of three algorithms: decision tree (DT), random forest (RF), and CatBoost.

The DT algorithm is on the Gini impurity splitting standard whereas, it deploys the weight of classes to account with the data imbalance, which helps in improving the accuracy of classification. In CIC IoT 2023, the DT simplified the model as ten salient characteristics through regression based on model vector.  On the whole, the CatBoost algorithm was superior to other ML-based approaches toward IDS and received the greatest average F1 score among all the datasets used. Such experiments show the ability of our proposed models to strengthen the IoT networks as far as security is concerned, even in resource constrained settings.