IDMM-IDS: An efficient and robust intrusion detection system for the IoT based on the inverted Dirichlet mixture model.

Abstract

The Internet of Things (IoT) has permeated all facets of modern life, offering revolutionary applications from smart homes to industrial automation. However, the widespread adoption of IoT systems has amplified security vulnerabilities, necessitating robust intrusion detection systems (IDSs) to protect these devices. Traditional IDS solutions often face challenges in resource-constrained IoT environments due to high computational demands and limited adaptability to emerging threats. To address these issues, this paper proposes IDMM-IDS, an efficient and robust IDS tailored for IoT contexts. By utilizing the inverted Dirichlet mixture model (IDMM) and extended stochastic variational inference (ESVI), our IDMM-IDS models complex network traffic with minimal computational overhead. Additionally, a novel cluster-based oversampling technique is integrated to address class imbalance, enhancing the detection of minority class threats without introducing noise. Extensive evaluations on three public datasets-UNSW-NB15, WSN-DS, and WUSTL-IIOT-2021-demonstrate that IDMM-IDS outperforms most existing methods in detection performance while significantly reducing training and decision times, making it well-suited for resource-constrained IoT environments.