Multi-class identification of driver's cognitive distraction with error-correcting output coding (ECOC) method

Human cognitive state monitoring is quite important for active safety which is a technology to prevent accidents. While driving, cognitive distraction like conversation or thoughts unrelated to driving has potential hazards. Identification of cognitive distraction has already been reported in literature. However, multi-class identification of cognitive distraction has yet to be reported. This paper suggests a multi-class identification with the Error-Correcting Output Coding (ECOC) method, which is one of the multi-class pattern recognition methods. Four sets of biological information are used as features for the identification of cognitive distraction. Two kinds of ECOC methods having different decoding rules were also compared. Consequently, higher identification capability of cognitive distraction was achieved. Finally, we discussed optimal combination of either two, three, or four feature sets and their implications.