<p>Worker fatigue is one of the most significant risks in hazardous industries such as construction. Traditional fatigue detection methods, which rely on subjective measures, are prone to bias and can interrupt work. Electroencephalography (EEG) signals have increasingly been used for the objective detection of fatigue, as they offer informative and rich data. However, their use has primarily been limited to detection of mental fatigue. This study examines whether EEG signals can support a unified fatigue detection model trained on both physical-task and mental-task segments. The study also further explores the performance of a fatigue detection model with a reduced number of EEG channels. To achieve these two objectives, a dataset consisting of objective EEG recordings combined with subjective fatigue surveys collected from 12 participants was used. The most important features were extracted using a systematic feature-selection approach. A Support Vector Machine (SVM) was then employed for the classification of physical and mental fatigue. To reduce the number of EEG channels, fifteen models representing different channel combinations were trained and tested using leave-one-subject-out cross-validation. The performance of each model was assessed using a custom performance score to determine the best combination of channels. The results showed that a two-channel model (TP9, AF7) achieved the best performance, with a score of 84. These findings provide important evidence for the development of purpose-built EEG devices with fewer channels capable of detecting both physical and mental fatigue.</p> Graphical abstract <p></p>