Self-Organizing Map (SOM) model for mental workload classification

Development of reliable mental workload measurement and classification techniques have been an area of concern in human factors engineering. Artificial Neural Networks (ANN) have been used to model workload by performing EEG data classification. In the present study, a Self-Organizing Map (SOM) neural network was used to simulate workload metrics, SOM is an unsupervised algorithm that clusters similar input vectors to allow its output neurons to compete among themselves to become activated. SOM functional features are considered to be similar to those of the human brain since the letter is capable of organizing heterogeneous sensory inputs. For purposes of this study, EEG data was preprocessed via Fast Fourier analysis, temporally segmented and reduced by Principal Component Analysis (PCA) prior to inputting it to the network. The network was trained using 2/3 of available data and tested with remaining 1/3 of the data to classify workload into six categories ranging from very low to overload The SOM was able to cluster the training data into 6 output categories and differentiate between workload classes when presented with the test data set. The results indicated that implementation of Se Organizing Map networks offers a robust method for analyzing electrophysiological data signals related to work performance and could potentially be used as a tool for extraction of workload correlates from EEG data. Knowledge about workload metrics and reliable classification methods can be utilized in the design of adaptive human-machine systems that control information flow to prevent operator overload.