Nowadays, more and more traffic research relies on plentiful intelligent transportation data. However, due to hardware, software or environmental reasons, traffic data may face various missing value problems. Previous studies have tried to tackle this problem using different models. However, as the missing rate increases, some models do not perform well. As we know, the GAN (generative adversarial network) model has achieved great success in the field of computer vision. This paper attempts to comprehensively impute the missing values at different missing rates by transforming the traffic data into image data. We use the modified GAN model to make the imputation. Studies have shown that the modified GAN model performs well compared with other models at different missing rates.

CICTP