Adaptive Truncated Schatten Norm for Traffic Data Imputation with Complex Missing Patterns

Accurate traffic data imputation is essential for Intelligent Transportation Systems (ITS), particularly under complex missing patterns. This study presents the Adaptive and Truncated Schatten Norm Low-Rank Tensor Completion (LRTC-ATSN) model, which employs the flexible Schatten norm for low-rank approximation and introduces an adaptive truncation mechanism to suppress noise and retain core features. To address non-convex optimization, the model integrates the Adan algorithm with Nesterov momentum, achieving equilibrium between computational efficiency and recovery precision through dynamic parameter tuning. To evaluate performance, a framework was designed to simulate diverse real-world traffic scenarios with mixed missing patterns. Extensive experiments on datasets from Guangzhou and Seattle demonstrate that LRTC-ATSN outperforms existing methods, yielding 10.6% lower MAPE and 6.1% lower RMSE relative to the best baseline model. Even with 95.85% data loss, the model maintains high reliability. These results underscore LRTC-ATSN’s potential for enhancing ITS data robustness and applications across domains like finance and healthcare.