Mental workload critically affects well-being and performance in safety-critical systems. While machine learning models for mental workload prediction often leverage physiological indicators, interpretability and error analysis are frequently overlooked. This study develops robust models for workload prediction that emphasize interpretability and analyzes common misclassifications to elucidate key mechanisms. Respiratory and cardiac signals from 30 participants, as well as oculomotor signals from 17 participants, captured under varying task demands were utilized. Five models of varying interpretability were validated with optimized hyperparameters and preprocessing. A logistic regression and a decision tree were selected to distinguish between two and three workload levels, respectively. On unseen test data, they achieved f1-scores of 90.5% (accuracy: 92.2%) and 72.0% (accuracy: 72.3%). Performance varied across scenarios and individuals. Findings show that transparent, efficient models combined with appropriate preprocessing can compete with black-box approaches, with implications for safety-critical applications where interpretability, trust, and computational efficiency are essential.

RASMUS ("Real-Time Assessment of Multidimensional User State") is a generic diagnostic framework designed to provide real-time diagnostics of user performance and up to six cognitive states impacting user performance. RASMUS diagnostics can be used in adaptive system designs to infer when the user needs support and to dynamically select and apply adaptation strategies that are suitable to restore the user's effectiveness in a given situation. In a proof-of-concept implementation, the generic RASMUS concept has been operationalized for a simulated air surveillance task. The implementation addresses three potentially problematic user states that are considered particularly relevant for this task domain: high workload, passive task-related fatigue, and incorrect attentional focus. This paper describes an experimental study aimed at validating the diagnostic capabilities of this proof-of-concept implementation. N=12 participants performed air-surveillance tasks in a scenario-based simulation. Congruent validity was assessed by comparing RASMUS diagnostics with subjective ratings of the respective user states. Ratings were obtained each time RASMUS detected a performance decrement. Outcomes showed that user states diagnosed as 'critical' by RASMUS were mostly consistent with subjective ratings at the time of performance decrements. However, a measure of Level 1-Situation Awareness proved to be more reliably associated with outcomes of RASMUS' "incorrect attentional focus" than a subjective measure of attention. Overall, outcomes support the validity of RASMUS diagnostics for the given task domain and the selected user states. Potential for further improvement is outlined.

User state assessment in adaptive human-computer-interaction has often been equaled to the assessment of workload. More recently, approaches have surfaced that also focus on other state dimensions such as fatigue, situation awareness, and the emotional state. However, interrelations and dependencies between these state dimensions are not considered if each is assessed in isolation. Furthermore, it is often neglected that individual factors, factors of the work environment, and self-regulation-strategies of the operator strongly influence these state dimensions. We suppose that adaptation based on user state assessment is more successful if this set of interactions is integrated into a more holistic assessment approach. We therefore propose a model that aims to provide a more holistic view on user state assessment. The model also aims to be generic in the sense that it can be applied to diverse operational conditions. Application is illustrated by a recent aviation incident.