The escalating complexity of technical systems has amplified cognitive demands on operators, potentially leading to increased mental fatigue. Traditional methods of managing fatigue, such as routine medical check-ups and strict control of working hours, have proven inflexible and inconsistent. Additionally, previous research relied on subjective measures to study the physiological changes shown by individuals after doing cognitive demanding tasks, which may not accurately represent the relationship between mental fatigue and physiological indicators. Instead, this study employs an objective measure of mental fatigue to investigate the relationship between mental performance and a range of physiological indicators estimated using deep learning models. Furthermore, this study employs machine learning techniques to detect mental fatigue through feature extraction, assessing the impact of including or excluding participant identification as a model input to explore individual variability in cognitive performance. Our findings indicate that detecting mental fatigue is a complex process that cannot rely solely on observable changes in facial expressions or eye features, as these manifestations differ across individuals. Thus, more advanced modeling approaches are necessary to uncover generalizable patterns. Supporting this, our experiment demonstrated that including participant identification as a feature improved the F1-score to 94%, compared to 81% when it was excluded, highlighting the significance of personal factors in accurately identifying mental fatigue.