Enhancing the cross-subject classification performance of EEG-based mental workload (MWL) monitoring models poses a significant challenge. Traditional methods require gathering calibration data for new users to prevent performance decline. However, the calibration data collection process is time-consuming and labor-intensive. In this study, we proposed a novel cross-subject MWL classification model that does not require calibration data. Specifically, we used periodic and aperiodic components obtained through EEG spectrum decomposition as features, replacing the commonly used power spectral density (PSD) features. These features are then aligned across subjects using a modified Euclidean alignment method. Our results show that the aligned periodic and aperiodic combined features achieve the highest classification accuracy (0.791±0.077), significantly surpassing raw PSD features without alignment (0.731±0.086, p<0.05). Moreover, we found a significantly negative correlation between inter-subject distances calculated from periodic features in resting-state data and inter-subject pairwise classification accuracy (r=-0.472, p<0.001). This finding suggests a promising approach to leverage resting-state data for selecting source subjects that closely match the target subjects.