Selective State-Space Neural Tucker Factorization for Temporal-aware Traffic Imputation
Abstract
Spatiotemporal traffic data imputation is a key task of intelligent transportation systems (ITS), and urban traffic data inherently exhibits strong long-term temporal dependencies. Therefore, accurate modeling of these long-term temporal dynamics is the core prerequisite for high fidelity traffic data imputation. However, existing traffic data imputation methods still suffer from inherent limitations in long-sequence modeling, such as gradient vanishing, low efficiency, and insufficient capability to capture long-range temporal correlations. To address these issues, this paper proposes a Selective State Space Neural Tucker Factorization model (S3NTF) for robust traffic data imputation. This model combines the Selective State Space mechanism with Neural Tucker Factorization to capture complex spatial correlations and long-term temporal dynamics. A state space-based time encoder is proposed to effectively extracts long sequence dependencies with linear complexity, while the latent spatiotemporal feature interaction learning module models high-order cross modal interactions based on Tucker decomposition. Extensive experiments on six real traffic datasets have shown that our model outperforms existing models in terms of accuracy.
Figures & tables
| No. | Dataset | Dimension | Known Count | Density (%) |
| D1 | Seattle City Speed | 323 28 288 | 260,467 | 9.9998 |
| D2 | New York City Flow | 30 30 1064 | 97,446 | 7.3957 |
| D3 | Hangzhou Flow | 80 105 28 | 20,976 | 8.9183 |
| D4 | Guangzhou Speed | 214 61 144 | 185,559 | 9.8713 |
| D5 | METR-LA Speed | 288 119 207 | 651,900 | 9.1891 |
| D6 | PEMS-BAY Speed | 288 181 325 | 1,693,717 | 9.9973 |
| Model | D1 | D2 | D3 | D4 | D5 | D6 | Win/Loss | Rank | ||||||||||||
| MAE | MRE | RMSE | MAE | MRE | RMSE | MAE | MRE | RMSE | MAE | MRE | RMSE | MAE | MRE | RMSE | MAE | MRE | RMSE | |||
| M1 | 7.3406 | 0.2282 | 10.5773 | 11.6017 | 4.2979 | 14.9034 | 70.7900 | 1.9060 | 123.2378 | 6.5716 | 0.2198 | 8.6559 | 8.9371 | 0.2377 | 12.1656 | 5.0974 | 0.1181 | 9.2495 | 0/18 | 8.00 |
| M2 | 5.4235 | 0.1356 | 7.8008 | 5.0889 | 0.6151 | 11.0201 | 25.2415 | 0.2217 | 59.7254 | 3.6128 | 0.1238 | 5.1250 | 6.0145 | 0.1482 | 8.8401 | 3.9895 | 0.0816 | 6.1690 | 0/18 | 6.17 |
| M3 | 5.4649 | 0.1371 | 7.8439 | 5.1034 | 0.6310 | 10.9229 | 25.2210 | 0.2210 | 58.9092 | 3.6610 | 0.1257 | 5.1598 | 5.9879 | 0.1476 | 8.8081 | 3.9887 | 0.0821 | 6.1943 | 0/18 | 6.28 |
| M4 | 4.9711 | 0.1293 | 7.2974 | 5.0559 | 0.6146 | 10.9956 | 24.9841 | 0.2425 | 55.1600 | 3.6676 | 0.1256 | 5.1623 | 5.7189 | 0.1426 | 8.6033 | 3.3521 | 0.0702 | 5.5257 | 0/18 | 5.39 |
| M5 | 4.5040 | 0.1148 | 6.7490 | 4.2742 | 0.5307 | 8.4263 | 24.9185 | 0.2175 | 55.8657 | 3.2746 | 0.1129 | 4.7575 | 4.7763 | 0.1184 | 7.5979 | 3.1842 | 0.0667 | 5.3216 | 0/18 | 3.83 |
Time Series Imputation
Time series imputation reconstructs missing observations in sequential data using temporal, cross-variable, and domain-specific patterns.
See all 67 papers