Temporal-aware Neural Tucker Factorization for Traffic Data Imputation

Le résumé fourni par la source

Modern transportation systems are integral to urban sustainability, economic vitality, and public safety. The effective analysis of spatiotemporal traffic patterns is critical for urban planning and smart city operations. However, real-world traffic monitoring system often produce high-dimensional and incomplete data, making accurate traffic modeling and control extremely challenging. This paper proposes a Gated-Recurrent Neural Tucker Factorization (GARNT) model for robust traffic data imputation. It integrates neural Tucker factorization with gated recurrent units (GRUs) to jointly model spatial correlations and temporal dynamics. A GRU-based temporal encoder captures temporal dependencies across time steps, while a novel latent spatiotemporal feature interactive learning module is responsible for modeling spatiotemporal interaction features based on the principle of Tucker Decomposition. Extensive experiments on six real-world traffic datasets demonstrates our model outperforms existing state-of-the-art methods in both accuracy and robustness. By combining the tensor decomposition’s interpretability with deep learning’s expressive power, GARNT offers a principled and scalable solution for data imputation tasks in intelligent transportation systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.