Unveiling Mental Workload via PPG: Morphological and Respiratory Feature–Driven Machine Learning Classification

The integration of the latest artificial intelligence (AI)-based technologies in high-risk operational sectors requires, as an essential prerequisite, the availability of reliable feedback on the mental workload (MWL) perceived by the operator. In this context, the analysis of variations in physiological signals remains one of the most promising and scalable approaches for estimating MWL in varied application scenarios. This study proposes the development of a predictive model solely based on the analysis of the photoplethysmographic (PPG) signal, a technology that is easily integrable into wearable systems and already widely used in clinical and consumer applications. In addition to the traditional variables associated with heart rate variability (HRV), which have been explored in previous literature, this work introduces an innovative analysis of morphological parameters of the individual pulse wave, which have not been previously investigated in the context of MWL. Furthermore, by reconstructing the respiratory contribution from the PPG signal, additional features related to respiratory variability were derived. The machine learning models were trained using the publicly available MAUS dataset, which includes recordings from 22 subjects exposed to controlled cognitive workload conditions through the N-back test. The obtained results, with an accuracy of 81.5% in the binary classification between low and high MWL levels, confirm the effectiveness of the proposed approach and highlight its potential for continuous, non-invasive monitoring of mental workload through a single wearable sensor.