Abstract
We investigated the appropriate range of time window for calculating eye and head movement parameters in mental workload (MWL) estimation during automobile driving. Participants performed driving tasks with a driving simulator, and eye and head movement were measured by controlling their MWL with the N-back task. The eye and head movement parameters were calculated by changing a time window from 30s to 150s in increments of 30s. An anomaly detector of MWL was constructed with one-class support vector machine (OCSVM) by using the data of no N-back task (“None”). In each window length condition, we calculated the area under curve (AUC) for the binary classification between None and the highest MWL condition, the percentage of anomaly data, and the distance from the decision boundary. The results showed that time window of 30s had significantly lower AUC than the other time windows. In addition, the correlation coefficient between the subjective MWL score and the distance from the decision boundary monotonically increased from 30s to 120s and decreased at 150s. Therefore, we concluded that 60s to 120s is an appropriate range as the time window for MWL evaluation.
