Table 2 Comparative performance of machine learning models (Mean ± Std).
From: Primacy of feature engineering over architectural complexity for intermittent demand forecasting
Model | MAE | RMSE | MAPE (%) | WMAPE (%) | \(\:{\varvec{R}}^{2}\) Score |
|---|---|---|---|---|---|
SHOS | 0.0712 ± 0.0028 | 1.0310 ± 0.1416 | 17.73 ± 0.47 | 20.39 ± 0.57 | 0.8430 ± 0.0460 |
Hurdle_SHOS | 0.0701 ± 0.0023 | 1.1136 ± 0.1102 | 16.84 ± 0.44 | 20.10 ± 0.47 | 0.8186 ± 0.0413 |
Hurdle_baseline | 0.1341 ± 0.0050 | 1.2911 ± 0.0934 | 37.91 ± 1.29 | 38.41 ± 0.71 | 0.7577 ± 0.0379 |
LightGBM | 0.1326 ± 0.0046 | 1.2900 ± 0.1079 | 38.25 ± 1.20 | 38.00 ± 0.68 | 0.7578 ± 0.0418 |
XGBoost | 0.1413 ± 0.0045 | 1.3960 ± 0.1074 | 40.41 ± 1.37 | 40.49 ± 1.26 | 0.7156 ± 0.0494 |
Random forest | 0.1364 ± 0.0046 | 1.3411 ± 0.1057 | 40.28 ± 1.54 | 39.08 ± 1.03 | 0.7377 ± 0.0484 |
Ridge | 0.3439 ± 0.0134 | 1.5101 ± 0.1161 | 69.99 ± 0.61 | 98.57 ± 4.28 | 0.6685 ± 0.0535 |
ElasticNet | 0.3439 ± 0.0134 | 1.5101 ± 0.1160 | 70.00 ± 0.61 | 98.59 ± 4.26 | 0.6685 ± 0.0535 |
Naive zero | 0.3492 ± 0.0137 | 2.6557 ± 0.0643 | - | 100.00 ± 0.00 | -0.0176 ± 0.0014 |
