Recognition of mental workload levels by combining adaptive exponential feature smoothing and locality preservation projection techniques

• Published in: Chinese Control Conference pp. 4700 - 4705
• Main Authors: Yin Zhong, Zhang Jianhua
• Format: Conference Proceeding
• Language: English
• Published: TCCT, CAA 01.07.2014
• Subjects: Accuracy; Electrocardiography; Electroencephalography; Electrooculography; exponential smoothing; Feature extraction; Indexes; locality preservation projection; mental stress; mental workload; operator functional state; Smoothing methods
• ISSN: 1934-1768
• DOI: 10.1109/ChiCC.2014.6895732

Assessing mental workload (MWL) in real time is crucial for preventing the accidents caused by cognitive overload or inattention of human operators in safety-critical human-machine (HM) systems. Since continuous-time psychophysiological signals reflect the mental stress of humans, classifiers designed by using those signals can be utilized to assess MWL effectively. However, the noise contained in the extracted temporal psychophysiological features may lead to the overlapping of classifications of different MWL levels at each time instant. In this paper, we tackle this problem by combining adaptive exponential smoothing (AES) of those high dimensional physiological features and locality preservation projection techniques (LPP) to improve the MWL classification performance. In a simulated HM-integrative process control system, the extracted psychophysiological features are first smoothed by AES scheme. Then, the dimensionality of the smoothed feature vector is reduced by using LPP to enhance the inter-class discrimination capacity. Based on the combination of AES and LPP techniques, the bias-added support vector machine (BSVM) is employed to realize the three-class (low, normal and high) MWL classification. It has been demonstrated that the proposed method can significantly improve the classification accuracy from 88.6% to 99.3% for subject-specific classifier design and from 79.3% to 93.6% for a generic classifier common to all individual subjects.