Cognitive fatigue is a transient, neurophysiologically grounded impairment of executive function and sustained attention arising from prolonged mental effort without adequate rest. It is a leading contributor to human error in safety-critical domains, including aviation, surgery, and long-haul transportation. Despite two decades of active research, deployed detection systems remain anchored to invasive physiological hardware, architecturally naive fusion strategies, and opaque, unexplained inference processes. These three barriers collectively prevent real-world deployment at any meaningful scale. This review systematically synthesises twenty landmark studies published between 2022 and 2026, drawn from IEEE Xplore, PubMed, Nature group journals, ACM Digital Library, and ScienceDirect. Three non-contact detection modalities are evaluated in depth: speech acoustic-prosodic features, facial Action Unit (AU) dynamics, and human-computer interaction (HCI) behavioural markers. The review also examines emerging multimodal fusion architectures and explainable artificial intelligence (XAI) integration across the literature. Six structural research gaps are formally identified: persistent reliance on invasive sensing hardware; temporally naive fusion that discards asynchronous cross-modal dependencies; the complete absence of HCI behaviour as a recognised primary detection modality; XAI confined to post-hoc unimodal attribution rather than embedded cross-modal fusion; inconsistent ground-truth labelling protocols that conflate cognitive fatigue with drowsiness; and the absence of any open, tri-modal benchmark dataset for community evaluation. Three primary research objectives are derived from this gap analysis, jointly defining the architecture and validation agenda for the proposed Explainable Multimodal Intelligence framework for Cognitive Fatigue Detection (EMI-CFD). This review provides a critical, evidence-grounded roadmap for the next generation of transparent, non-invasive cognitive fatigue monitoring systems.

Keywords: cognitive fatigue detection; multimodal fusion; explainable artificial intelligence; speech prosody; facial action units