The objective of this study was to apply diverse machine learning algorithms to differentiate low and high MWL levels using a variety of psychophysiological signals (obtained from fNIRS, ECG, and pupillometry). In this study, we assessed various machine learning classifiers, including LDA, Decision Tree, Logistic Regression, Random Forest, Multilayer Perceptron, kNN, SVM, and Naïve Bayes. Classification analysis utilized PyCaret, a Python library that facilitates model training and assessment. The findings indicated that models utilizing only pupil data attained the highest accuracy (75%), while integration with additional physiological metrics like fNIRS and ECG resulted in accuracy surpassing 70%, particularly with linear discriminant analysis and decision tree methods. The integration of various physiological signals, notably pupil data, significantly enhances the precision of mental workload classification. Conversely, models relying on individual physiological signals exhibited lower accuracy, with results ranging from 38% to 64%. These results underscore the significance of integrating multiple inputs and selecting appropriate classifiers to optimize mental workload detection efficacy.