Driver identification using in-vehicle digital data in the forensic context of a hit and run accident

Abstract One major focus in forensics is the identification of individuals based on different kinds of evidence found at a crime scene and in the digital domain. In the present study, we assessed the potential of using in-vehicle digital data to capture the natural driving behavior of individuals in order to identify them. Freely available data was used to classify drivers by their natural driving behavior. We formulated a forensic scenario of a hit and run car accident with three known suspects. Suggestions are provided for an understandable and useful reporting of model results in the light of the requirements in digital forensics. Specific aims of this study were 1) to develop a workflow for driver identification in digital forensics, 2) to apply a simple but sound method for model validation with time series data and 3) to transfer the model results to answers to the two forensic questions a) to which suspect does the evidence most likely belong to and b) how certain is the evidence claim. Based on freely available data (Kwak et al., 2017) the first question could be answered by unsupervised classification using a random forest model validated by random block splitting. To answer the second question we used model accuracy and false detection rate (FDR) which were 93% and 7%, respectively. Furthermore, we reported the random match probability (RMP) as well as the opportunity of a visual interpretation of the prediction on the time series for the evidence data in our hypothetical hit and run accident.