Synthetic CANBUS Data Generation for Driver Behavior Modeling

Ucuzova, Esranur
Kurtulmaz, Ekim
Gökalp Yavuz, Fulya
Şahin, Nuri Eray
It aims to develop an artificial intelligence model that can analyze driver behavior and compare the model's performance between real data and synthetic data by testing this model on different synthetic data. In general, a huge amount of data is needed to develop a successful artificial intelligence model; especially, due to reasons such as the difficulty of obtaining the necessary data to model driver behavior, ways of synthesizing new data with the real data at hand were investigated. Accordingly, the synthpop library, which can create an entirely new data set from real data by maintaining the basic statistics and distribution of the data, and doing this with the CART algorithm, was used. The synthetic data set obtained is tested on an artificial intelligence model that performs driver behavior analysis trained with real data; test results of both data were compared and, as a result, promising results were obtained. Accordingly, it has been concluded that data from different fields, especially in areas where it is difficult to obtain data such as vehicle usage data, can be used to increase the performance of existing models by reproducing with the synthpop library.
Citation Formats
E. Ucuzova, E. Kurtulmaz, F. Gökalp Yavuz, H. KARACAN, and N. E. Şahin, “Synthetic CANBUS Data Generation for Driver Behavior Modeling,” presented at the IEEE 29th Signal Processing and Communications Applications Conference (SIU), Türkiye, 2021, Accessed: 00, 2021. [Online]. Available: