Synthetic data generation using Copula model and driving behavior analysis

Abstract

In this study, the generation of synthetic driving data that can reflect real behavior well using the Copula model was investigated. To see the difference in behavior patterns in the generated synthetic driving data, a feature correlation comparison was made. The difference in driving behavior was provided with the K-means based classification model. It was shown that with a Random Forest model trained with synthetic data and having high accuracy, the privacy of real data could be protected by 98.55%. At the end of the study, it was seen that the Copula model could obtain synthetic driving data with sufficient accuracy with CAN bus data without additional sensor support.