Identifying Deviations from Normal Driving Behavior

University Transportation Centers Program2022PDFTech ReportHananeh Alambeigi, HMcDonald, Anthony DShipp, EvaManser, MichaelSafety through Disruption (Safe-D) University Transportation Center (UTC)Texas A&M Transportation InstituteUnited States. Department of Transportation. Office of the Assistant Secretary for Research and TechnologyUnited States. Department of Transportation. University Transportation Centers (UTC) ProgramSafety through Disruption (Safe-D) University Transportation Center (UTC)United StatesBehaviorData collectionData filesDriver performanceDriver support systemsDriversDrivingErrorsPredictive modelsMachine learningDriver BehaviorAutomated DrivingTransfer of ControlPhysiological MeasuresTTI-Student-0869A3551747115/TTI-Student-08Final ReportOne of the critical circumstances in automated vehicle driving is transition of control between partially automated vehicles and drivers. One approach to enhancing the design of transition of control is to predict driver behavior during a takeover by analyzing a driver\u2019s state before the takeover occurs. Although there is a wealth of existing driver behavior model prediction literature, little is known regarding takeover performance prediction (e.g., driver error) and its underlying data structure (e.g., window size). Thus, the goal of this study is to predict driver error after a takeover event using supervised machine learning algorithms on various window sizes. Three machine learning algorithms\u2014decision tree, random forest, and support vector machine with a radial basis kernel\u2014were applied to driving performance, physiological, and glance data from a driving simulator experiment examining automated vehicle driving. The results showed that a random forest algorithm with an area under the receiver operating curve of 0.72, trained on a 3 s window before the takeover time, had the highest performance for accurately classifying driver error. In addition, we identified the 10 most critical predictors that resulted in the best error prediction performance. The results of this study can be beneficial for developing driver state algorithms that could be integrated into automated driving systems.127