Di-GraphGAN: An enhanced adversarial learning framework for accurate spatial-temporal traffic forecasting under data missing scenarios

Published: 01 Jan 2024, Last Modified: 30 Sept 2024Inf. Sci. 2024EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•We propose Di-GraphGAN, an enhanced framework for end-to-end traffic data imputation and prediction in various missing scenarios.•DI-LSTM, an RNN-based data imputation module with Time Damping unit, is proposed to impute missing traffic values and enhance prediction accuracy.•We introduce TE-GAT, which employs Filter Gate to retain only important edges for neighbor aggregation, reducing the computational overhead.•Comprehensive experiments on datasets from Hangzhou and Guangzhou show Di-GraphGAN consistently outperforms benchmarks.
Loading