URBAN TRAFFIC FLOW PREDICTION USING TIMEGAN MODEL

The gap between the supply and demand for road space has been widening over the past few decades as the number of cars has increased steadily. This imbalance has resulted in severe traffic congestion, hindering the development of the social economy. Accurate traffic flow forecasting becomes crucial to address this issue, as it enables effective management of factors impacting traffic flow, such as accidents, congestion, special events, and road closures. While several models have been proposed, achieving an accuracy level between 85% to 88%, this article introduces a novel methodology called the time-series generative adversarial network (TimeGAN) model for improved traffic flow prediction. In several areas, including stock prediction, image classification, creative image creation, and synthesis of sequential data, the TimeGAN model has demonstrated excellent performance. TimeGAN is used in this work to investigate a more successful method of predicting traffic flow. For validating the effectiveness of the proposed work, four prediction models, such as multiple linear regression, polynomial regression, LSTM (Long Short-Term Memory), and bidirectional LSTM, are also employed for an extensive comparative analysis. The results demonstrate that TimeGAN significantly outperforms the other models, achieving an accuracy of 90% for a 10-minute prediction window. This highlights the superiority of the TimeGAN model in accurately forecasting traffic flow compared to the alternative approaches considered in the study.