부산대학교 도서관

In this study, a method combining statistical and machine learning approaches is proposed to select the most informative EEG features for the detection of the cognitive load linked to fine motor activities, during a Purdue Pegboard Test (PPT). The proposed method is validated by an experimental case study on neurosurgeons monitored by means of a wearable electroencephalographic (EEG) acquisition system during PPT execution at four increasing levels of complexity. EEG features of cognitive workload related to fine motor activity are identified by a Speraman rank correlation analysis and Friedman test. The EEG features selected by the statistical approach are provided as input to the machine learning algorithms. Classification accuracy is the metric adopted to validate the results of feature selection. Among the tested classifiers, the k-Nearest Neighbor (k-NN) reaches 53.3 ± 4.5% average accuracy in detecting the four levels of cognitive load.