Research Article
DWNet: Dual-Window Deep Neural Network for Time Series Prediction
Table 2
MAPE and SMAPE performance comparison among different methods and datasets (best result is displayed in boldface).
| Models | SML 2010 | Human Sports | EEG | Energy | MAPE (%) | SMAPE (%) | MAPE (%) | SMAPE (%) | MAPE (%) | SMAPE (%) | MAPE (%) | SMAPE (%) |
| ARIMA (16) | 123.0993 | 62.0098 | 22.3507 | 18.4392 | 159.6348 | 83.7905 | 178.4365 | 77.4287 | LSTM (16) | 78.9095 | 45.0082 | 17.4439 | 13.9803 | 80.3427 | 56.9819 | 163.9849 | 65.8066 | LSTM (128) | 83.1021 | 49.5439 | 17.9035 | 12.0033 | 87.5609 | 63.5271 | 176.3415 | 69.5442 | Encoder-Decoder (16) | 70.7142 | 43.0760 | 13.3326 | 9.5610 | 66.4635 | 38.5606 | 170.0863 | 69.3110 | Encoder-Decoder (128) | 78.9981 | 50.5022 | 15.7987 | 10.0923 | 79.3445 | 40.2327 | 181.7583 | 76.1764 | Input-Attn-RNN (16) | 61.1121 | 30.0997 | 11.7831 | 7.9462 | 41.8856 | 32.4032 | 145.8688 | 67.5386 | Input-Attn-RNN (128) | 68.9089 | 35.3459 | 12.0034 | 7.9897 | 41.4628 | 29.7608 | 152.3287 | 64.7685 | Temp-Attn-RNN (16) | 54.3435 | 32.8703 | 11.2627 | 7.2110 | 40.8683 | 29.0871 | 140.6838 | 56.0774 | Temp-Attn-RNN (128) | 57.8065 | 31.9911 | 11.4980 | 7.1153 | 45.8705 | 30.0085 | 128.0644 | 59.3527 | TCN (16) | 83.2350 | 49.5797 | 18.5920 | 11.0097 | 675.9030 | 131.4202 | 258.1707 | 93.9882 | TCN (128) | 85.4479 | 67.3689 | 14.6141 | 10.7326 | 995.0083 | 133.4580 | 265.3023 | 100.9782 | LSTNet (16) | 50.5956 | 29.6186 | 13.0975 | 8.6524 | 46.9208 | 34.3753 | 135.8396 | 68.8810 | LSTNet (128) | 83.2999 | 46.3060 | 13.3192 | 9.3698 | 50.0573 | 41.5482 | 140.2021 | 72.9974 | DARNN (16) | 43.1275 | 28.9558 | 11.9568 | 8.0686 | 36.4658 | 26.6514 | 123.0556 | 59.8798 | DARNN (128) | 45.2110 | 33.1612 | 11.8951 | 7.4177 | 33.8550 | 27.1255 | 139.0686 | 64.3326 | DSTP (16) | 40.6946 | 24.5261 | 11.7359 | 7.1343 | 34.5063 | 22.7594 | 138.8959 | 59.8884 | DSTP (128) | 36.2600 | 24.8048 | 11.7928 | 7.3406 | 35.1179 | 24.0093 | 142.8744 | 56.9903 | DWNet | 31.5764 | 23.0888 | 10.3833 | 7.0483 | 31.3287 | 20.6706 | 82.1119 | 52.5880 |
|
|
The window size of baseline methods is set to 16 and 128, and the short window size and long window size of DWNet are set to 16 and 128, respectively.
|