Online Assessment of Spontaneous Mental Fatigue in Construction Workers Considering Data Quality: Improved Online Sequential Extreme Learning Machine

Biological data-based methods for monitoring workers’ mental fatigue have become widely adopted in recent years. However, few have concentrated on the online monitoring and assessment of mental fatigue considering the complexity and high dimension of the biological data, especially for scenarios where data arrives continuously in the form of flows. This study aimed to propose an online learning model to learn model parameters according to the order of data acquisition. Specifically, the fuzziness-based online sequential extreme learning machine (Fuzziness-OS-ELM) model was proposed, consisting of two parts: (1) a data value estimator; and (2) an online mental fatigue classification model. As new data arrives, the Fuzziness-OS-ELM model can effectively identify and select samples with high data quality based on fuzziness, which are then used to continuously update the online mental fatigue classification model. A cognitive experiment was carried out to evaluate the Fuzziness-OS-ELM model. The results indicated that samples with low fuzziness corresponded to high data quality. The proposed online sequential learning model exhibited enhanced classification performance on mental fatigue. This study’s dynamic diagnostic method for identifying the onset and progression of mental fatigue can provide targeted support for precise interventions aimed at construction workers.