Non-invasive physical demand assessment using wearable respiration sensor and random forest classifier

Physically demanding tasks are one of the leading causes of fatigue among workers in labor-intensive industries such as construction. Despite the recent development in physical demand assessment, there is a lack of a practical solution to reduce injuries and illnesses resulted from high-intensity tasks. Consequently, this study proposes a framework to assess the workers' physical demand level using an off-the-shelf respiration sensor. In the proposed framework, the extracted features from respiratory signals in the time and frequency domain are used to train the random forest classifier. Then, the trained model is used to classify the physical demand of the worker in new observations. To evaluate the performance of the proposed framework, an experiment including a masonry wall construction task was designed where the respiratory signals of the 15 participants were recorded. Then, the collected signals were labeled using the NASA-TLX questionnaire. The results showed that the proposed framework increases the accuracy of physical demand classification up to 93.4% while being less sensitive to body and sensor movement artifacts. Moreover, physical demand assessment was performed using a single bio-signal while eliminating the need for monitoring multiple bio-signals simultaneously. The findings of this study should make an important contribution to workers' safety, well-being through the detection of high physical load on workers.