Suitability of Machine Learning and Deep Learning Algorithms in IIoT Attack Detection
Prathibha Keshavamurthy, Erin Colleen Smith, Aasif Majeed, Sarvesh S. Kulkarni · 2025
Industrial Internet of Things (IIoT) devices improve the quality of data and technology in the healthcare, commercial, industrial, and transportation realms. However, with their interconnectedness come many security risks to these industries. Many complex solutions involving Machine Learning (ML), Deep Learning (DL) or an ensemble of algorithms have been proposed for identifying malicious packets targeting IIoT devices. However, increasing complexity leads to increases in model training and detection times. Therefore, a balance between model accuracy and its computation time is advisable. In this paper, we analyze the performance of ML and DL algorithms on the Edge IIoT dataset in detecting attacks on IIoT devices. We use a total of 108 models to evaluate six ML and six DL algorithms in their classification of packets into normal or attack packets across binary, multigroup, and multiclass classifications. We demonstrate that ML algorithms perform equally well and in some cases marginally better than DL algorithms, and with considerably less computational overhead. Therefore, the ML algorithms are better-suited for IIoT-related implementations. Furthermore, we demonstrate that a reduced feature set reduces computation time without appreciably degrading detection metrics.