Mental fatigue arises from prolonged cognitive activity and is typically associated with cognitive decline and feelings of tiredness and demotivation. Machine learning offers a promising approach for identifying fatigue in individuals, which can help prevent its negative consequences. Previous research has primarily trained algorithms on EEG data, whereas other physiological indicators have been used less frequently. In this study, fatigue was induced by having participants (n = 28) complete nine blocks (~30 minutes total) of a visually guided pointing task. Cardiac activity, eye movements, pupil diameter, and skin conductance were recorded throughout the task. Data from the first two blocks were labelled “non-fatigued,” and data from the last two blocks were labelled “fatigued.” Fatigue was identified using supervised classification with leave-one-subject-out cross-validation. As expected, cognitive performance declined and subjective fatigue increased over time. Physiological measures also changed significantly as a function of time-on-task. The random forest classifier achieved an accuracy of 79.46%. Feature selection revealed that skin conductance parameters were the strongest predictors, followed by heart rate variability, fixation instability, and phasic pupil-size changes. These cognitive and physiological changes are in line with previous evidence of motor-cognitive deficits and increased parasympathetic activity under fatigue. Although classification accuracy is comparable to studies using peripheral physiological signals, it remains markedly lower than the ~90% typically achieved with EEG-based approaches.