<p>The Industrial Control System (ICS) and isolated air-gapped networks are critical to modern manufacturing and highly sensitive infrastructure. While these systems are intentionally isolated from unsecured networks, they remain vulnerable to physical side-channel attack vectors. In recent years, Electromagnetic (EM) side-channel analysis has emerged as a powerful tool in studying both the exfiltration of data and the forensic detection of anomalous system behaviors. However, what is not yet clear is the effect of concurrent background processes and heavy system workloads on the reliability of these detection mechanisms. Most studies in the field of EM side-channel analysis have only focused on binary classification tasks (e.g., idle vs. malicious), which fail to estimate the true rate of false-positives generated by overlapping electromagnetic interference (EMI) in multi-core processors. The principal objective of this project was to investigate the operational limitations of current EM methodologies and propose a conceptual theoretical framework based on multi-class baselines. A systematic literature review was conducted in accordance with Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines, resulting in the rigorous synthesis of 52 pure-EM studies. The findings clearly indicate that current binary classifiers catastrophically degrade in accuracy when subjected to mathematically intensive Central Processing Unit (CPU) noise and routine operational loads. These findings provide a solid evidence base for the necessity of transitioning from binary models to a novel 3-class (Idle, Benign-Heavy, Malicious) machine learning detection framework.</p>