Table D1
Configuration of the ML model. (1) refers to the error value for 1-day forecasting. Same for (2) refers to 2-day forecasting, and (3) for 3-day forecasting. *In the 1D-CNN layer, 32 filters, a kernel size of 5, and strides of 1 were used.
| Model architecture | No. of hidden layers | No. of hidden neurons | Activation function | Batch size | Learning rate | Epochs | Callbacks functions | Validation Set |
Testing Set |
||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| MAE | MSE | RMSE | MAPE | MAE | MSE | RMSE | MAPE | ||||||||
| Linear | – | – | – | 64 | 0.001 | 100 | EarlyStopping | 0.312 | 0.141 | 0.376 | 87.883 | 0.143 | 0.045 | 0.211 | 60.689 |
| Dense ML | 2 | 32 | ReLU | 64 | 0.001 | 100 | EarlyStopping | 0.262 (1) | 0.118 (1) | 0.344 (1) | 132.580 (1) | 0.400 (1) | 0.281 (1) | 0.530 (1) | 238.898 (1) |
| 0.275 (2) | 0.138 (2) | 0.372 (2) | 132.004 (2) | 0.395 (2) | 0.286 (2) | 0.535 (2) | 234.704 (2) | ||||||||
| 0.290 (3) | 0.166 (3) | 0.407 (3) | 129.288 (3) | 0.392 (3) | 0.294 (3) | 0.542 (3) | 230.896 (3) | ||||||||
| Simple RNN | 2 | 32 | Tanh | 64 | 0.001 | EarlyStopping | 0.143 (1) | 0.035 (1) | 0.187 (1) | 70.990 (1) | 0.178 (1) | 0.052 (1) | 0.228 (1) | 69.624 (1) | |
| 100 | ModelCheckpoint | 0.171 (2) | 0.063 (2) | 0.251 (2) | 68.694 (2) | 0.171 (2) | 0.071 (2) | 0.266 (2) | 78.075 (2) | ||||||
| 0.264 (3) | 0.118 (3) | 0.343 (3) | 72.505 (3) | 0.200 (3) | 0.084 (3) | 0.289 (3) | 67.416 (3) | ||||||||
| Stateful RNN | 3 | 32 | Tanh | 64 | 1.58e−4 | 100 | LearningRateScheduler | 0.203 (1) | 0.060 (1) | 0.244 (1) | 56.390 (1) | 0.155 (1) | 0.039 (1) | 0.197 (1) | 59.988 (1) |
| EarlyStopping | 0.305 (2) | 0.131 (2) | 0.362 (2) | 81.028 (2) | 0.223 (2) | 0.079 (2) | 0.281 (2) | 71.679 (2) | |||||||
| 0.349 (3) | 0.173 (3) | 0.416 (3) | 82.819 (3) | 0.223 (3) | 0.084 (3) | 0.289 (3) | 64.159 (3) | ||||||||
| Stateful LSTM | 3 | 32 | Tanh | 64 | 1.58e−4 | 100 | EarlyStopping | 0.095 (1) | 0.021 (1) | 0.146 (1) | 40.335 (1) | 0.098 (1) | 0.020 (1) | 0.141 (1) | 41.781 (1) |
| 0.151 (2) | 0.048 (2) | 0.220 (2) | 48.937 (2) | 0.134 (2) | 0.042 (2) | 0.205 (2) | 57.860 (2) | ||||||||
| 0.174 (3) | 0.076 (3) | 0.275 (3) | 55.662 (3) | 0.166 (3) | 0.071 (3) | 0.267 (3) | 68.025 (3) | ||||||||
| Stateful Bi-LSTM | 3 | 32 | Tanh | 64 | 1.58e−4 | 100 | EarlyStopping | 0.149 (1) | 0.043 (1) | 0.207 (1) | 58.151 (1) | 0.170 (1) | 0.049 (1) | 0.221 (1) | 71.059 (1) |
| 0.190 (2) 0.249 (3) | 0.074 (2) 0.120 (3) | 0.272 (2) 0.347 (3) | 60.154 (2) 67.988 (3) | 0.211 (2) 0.229 (3) | 0.090 (2) 0.108 (3) | 0.300 (2) 0.329 (3) | 92.727 (2) 87.049 (3) | ||||||||
| 1D-CNN LSTM | 3 | 32 (5,1)* | ReLU | 64 | 1.58e-4 | 100 | EarlyStopping | 0.108 (1) | 0.027 (1) | 0.165 (1) | 41.164 (1) | 0.098 (1) | 0.023 (1) | 0.151 (1) | 51.732 (1) |
| 0.146 (2) | 0.051 (2) | 0.226 (2) | 47.512 (2) | 0.138 (2) | 0.047 (2) | 0.217 (2) | 68.376 (2) | ||||||||
| 0.177 (3) | 0.078 (3) | 0.279 (3) | 53.087 (3) | 0.156 (3) | 0.067 (3) | 0.259 (3) | 69.338 (3) | ||||||||
Current usage metrics show cumulative count of Article Views (full-text article views including HTML views, PDF and ePub downloads, according to the available data) and Abstracts Views on Vision4Press platform.
Data correspond to usage on the plateform after 2015. The current usage metrics is available 48-96 hours after online publication and is updated daily on week days.
Initial download of the metrics may take a while.