Predicting Types of Drivers



Journal Title

Journal ISSN

Volume Title



Driver types and their associated behaviors not only shape our driving habits but our reactions in unintended driving events as well. The open road places us in unexpected situations and forces us to act and react in particular ways. Here, in this thesis, I pro- pose a method to predict the types of drivers and their associated reactions during unintended events. We demonstrate our clustering and predicting methods with data from two simulations, On-road Driving (ORD) and Test Track Driving 1 (TTD1). In On-road Driving Study (n = 8), we construct a between-variable predicting model to predict the level of arousal of perinasal perspiration for the next 5 seconds based on driving variables of the last 30 seconds. Subsequently, we use TTD1 (n = 21) data to de- velop a within-variable model to predict the arousal of drivers during an unintended acceleration event based on their arousal levels in driving tests simulating our daily driving. We achieve a classification performance AUC at 0.96 and 0.90 for between- variable prediction model and within-variable predicting model, respectively. We also find a group of accelerophobic drivers whose stress level increases along with the ac- celeration of vehicles. The proposed method can also be used in the design of future vehicles; the types of drivers could be detected and embedded in advanced automation systems to personalize car driving variables or enhance car safety features accordingly.



Affective computing, driving behavior, sudden unintended acceleration, machine learning, extreme gradient boosting