Application of Multi-Criteria Decision Analysis for Quantifying Responsibility in ML-based Intrusion Detection System

With the swift growth in Internet of Things (IoT), certifying secure and trustworthy networks has turned to be a critical challenge, particularly as IoT devices are increasingly vulnerable to sophisticated cyberattacks. As a remedy, intelligent intrusion detection systems (IDS) evolved as promising solutions in recent years, but deciding on the appropriate model remains difficult because of competing performance and trustworthiness criteria. To this end, this paper explores a novel application of an ML-augmented decision-making framework to enhance security-related decision-making in IoT environments. The framework systematically evaluates and ranks ML-based IDS systems according to different evaluation criteria with distinct trade-offs, including detection performance, computational efficiency, algorithmic fairness, and model explainability. The ranking and decision are conducted through the application of the TOPSIS (Technique for Order Preference by Similarity to Ideal Solution) method, and other competitors. To demonstrate its applicability, we conduct experiments on two widely used IoT security datasets, namely NSL-KDD and Edge-IIoT. The experimental results show that the Random Forest (RF) IDS achieves superior performance on the NSL-KDD dataset. In contrast, the Extra Trees (ET) classifier emerges as the most suitable model for the Edge-IIoT dataset, offering a balanced trade-off among accuracy, training efficiency, low bias, and high explainability. These results demonstrate the applicability and effectiveness of our framework in guiding the selection of trustworthy and responsible IDS for diverse IoT security scenarios