A Comprehensive Survey on Implementation Challenges of Machine Learning-Based Intrusion Detection Systems in Operational Industrial Environments

A Comprehensive Survey on Implementation Challenges of Machine Learning-Based Intrusion Detection Systems in Operational Industrial Environments

Ali Amirpoorsaeed1 Zeinab Kalantari2

1) Department of Computer, Raf.C. , Islamic Azad University, Rafsanjan , Iran
2) Department of Computer, Raf.C. , Islamic Azad University, Rafsanjan , Iran Email:

IOI Code : AAHD-ABKHKA
Publication : National Conference on Information Technology, Nanotechnology, Artificial Intelligence and Technological Futures Studies(itnaf.ir)
Abstract :
    With the increasing connectivity of Industrial Control Systems (ICS) and Supervisory Control and Data Acquisition (SCADA) systems to corporate networks and the Internet, Machine Learning (ML)-based Intrusion Detection Systems (IDSs) have emerged as a promising solution for monitoring the security of industrial environments. However, transferring these models from laboratory settings to real-world industrial operations involves significant challenges. This survey reviews the implementation challenges of ML-based IDSs in operational industrial environments, ranging from data preparation to deployment on the factory floor. The review covers English-language studies published between 2018 and 2026 in IEEE Xplore, ScienceDirect, SpringerLink, MDPI, and arXiv, with a focus on challenges related to data, modeling, deployment in Operational Technology (OT), and operational evaluation. Following a structured screening process, nine key sources were identified and analyzed in depth. The findings indicate that the main barriers include the scarcity of real labeled attack data, class imbalance, difficulty in detecting zero-day attacks, hardware constraints on edge devices, high false-positive rates in noisy environments, and limited model generalizability across different sites. Moreover, reliance on unrealistic evaluation metrics, particularly Accuracy alone, together with the absence of standardized benchmarks, makes fair comparison difficult. Finally, future research directions are proposed, including more realistic open benchmarks, operational evaluation frameworks, integration of OT domain knowledge with ML, and MLOps for industrial IDSs.

Keywords : industrial control systems (ics) scada intrusion detection systems (ids) machine learning (ml) industrial cybersecurity it/ot convergence