EEG Based Driver Inattention Identification via Feature Profiling and Dimensionality Reduction

More than 90% of the persistently increasing traffic fatalities is related to human choice/error. Monitoring driver attention has a direct effect on decreasing injury/fatality rates. In recent years, there has been much effort to estimate drivers’ state with the goal of improving their driving behavior and preventing vehicle crashes in the first place. Physiological based detection has shown to be the most direct method of measuring driver state among which, electroencephalogram (EEG) is the most comprehensive method. EEGs are recorded from multiple channels that are processed separately. However, contribution of a fairly large number of the channels might be minimal to the target application. The computational load and the redundancy induced by those channels can hurt the identification performance. In this study, we propose an EEG-based systematic methodology for the assessment of driver state of inattention. Our proposed framework includes three major modules: (1) We first characterize each EEG channel rigorously via extraction of various categories of descriptors as features, (2) we then capture the contribution of each channel toward the identification task via channel specific feature dimensionality reduction, (3) we then conduct channel selection in order to find key brain regions of impact. Eight subjects participated in our naturalistic driving study. Our proposed method resulted in the accuracy of 98.99 ± 1.2% inattention identification accuracy. We also discovered that the first and second best channels are consistently selected from frontal and parietal regions for participating subjects.