Table 4 Comparison of fusion strategies for combining \(\operatorname {WTC}\) and \(\operatorname {FPA}\) outputs on METR-LA dataset. \(\operatorname {UB}\) denotes Uncertainty-Based fusion. \(\operatorname {LA}\) represents Learned Attention. Our hybrid approach combines both mechanisms.
From: Multi-scale Wavelet-Mamba framework for spatiotemporal traffic forecasting
Fusion Strategy | 15-min | 30-min | 60-min | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
\(\operatorname {MAE}\downarrow\) | \(\operatorname {RMSE}\downarrow\) | \(\operatorname {MAPE}\downarrow\) | \(\operatorname {Avg.}\) | \(\operatorname {MAE}\downarrow\) | \(\operatorname {RMSE}\downarrow\) | \(\operatorname {MAPE}\downarrow\) | \(\operatorname {Avg.}\) | \(\operatorname {MAE}\downarrow\) | \(\operatorname {RMSE}\downarrow\) | \(\operatorname {MAPE}\downarrow\) | \(\operatorname {Avg.}\) | |
Equal Weighting | 2.84 | 5.97 | 7.89 | 5.57 | 3.42 | 7.23 | 9.67 | 6.77 | 4.12 | 8.45 | 11.23 | 7.93 |
LA Only | 2.79 | 5.89 | 7.76 | 5.48 | 3.36 | 7.15 | 9.54 | 6.68 | 4.05 | 8.37 | 11.08 | 7.83 |
UB Only | 2.76 | 5.85 | 7.68 | 5.43 | 3.31 | 7.09 | 9.45 | 6.62 | 3.98 | 8.29 | 10.95 | 7.74 |
Hybrid (Ours) | 2.73 | 5.81 | 7.63 | 5.39 | 3.28 | 6.97 | 9.43 | 6.56 | 3.94 | 8.23 | 10.89 | 7.69 |
