Detection of Trojan Horse in the Internet of Things: Comparative Evaluation of Machine Learning Approaches

Abstract

As the Internet of Things (IoT) continues to expand, integrating interconnected devices into daily life, the threat landscape similarly evolves with increasing cybercrime activities. Among these threats, Trojan horse malware poses a significant risk to IoT systems by disguising itself as legitimate software, thereby gaining unauthorized access and control over networked devices. These intrusions are particularly challenging to detect in IoT environments due to the diverse and often less secure nature of connected devices, allowing malicious activities to persist undetected, leading to data breaches, device malfunction, or unauthorized data transmission to attackers. Addressing the inadequacies of traditional malware detection methods, this paper introduces a novel machine learning-based approach tailored to the IoT context. Leveraging a detailed dataset of 177,482 instances across 86 attributes, we evaluate the efficacy of various algorithms, including Naive Bayes (NB), J48 trees, and IBk to detect Trojan horse activities specifically in IoT devices. Our research comprehensively compares these methods, aiming to enhance detection accuracy and adaptability in dynamic IoT settings significantly.