Generating Realistic Synthetic Traffic Data using Conditional Tabular Generative Adversarial Networks for Intelligent Transportation Systems

Intelligent transportation systems (ITS) are a set of technologies that can be used to improve the efficiency, safety, and sustainability of transportation systems. One of the key challenges in ITS is data sparsity. One of the main reasons for data sparsity in ITS is the high cost of maintaining and deploying ITS infrastructure. To overcome the data sparsity is-sue, researchers have proposed different approaches to generate synthetic traffic data, such as employing traffic simulators or statistical methods. Generative Adversarial Networks (GANs) are a powerful class of deep learning models that excel in generating synthetic data that exhibits the characteristics and found patterns in the training data, particularly in the domains of images and text. In this paper, we discuss the data sparsity challenge in ITS and present our pioneering use of Conditional Tabular Generative Adversarial Networks (CTGANs) to generate synthetic traffic data. It takes into account the conditional dependencies among columns in the dataset, enabling the generation of samples that respect the relationships and patterns observed in the real data. In the field of ITS, the application of CTGANs is unexplored. While CTGANs have been used in other domains such as intrusion detection systems and network traffic analysis, their utilization in ITS is novel and unprecedented. CTGANs offer the potential to generate fine-grained and realistic data, making them valuable for creating accurate simulation setups in ITS. We evaluate our approach using a range of real-world traffic datasets and demonstrate that our method produces realistic and accurate synthetic traffic data. Additionally, we assess the generalization capability of our synthetic data, finding that it exhibits strong performance on unseen data.