Research Article
An Approach for Demand Forecasting in Steel Industries Using Ensemble Learning
Table 5
Best
-performing ML model and preprocessing and tuned hyperparameters with the highest possible accuracy
.
| ML techniques | Best preprocessing | Best hyperparameters | Performance |
| SVR | Processed data Corr (n_attributes: 4) | | | | Kernel: RBF | | | Epsilon: 0.1 | | | Gamma: 0.001 | |
| RFR | Processed data PCA (n_attributes: 4) | n_estimators: 10 | | | Criterion: mse | | | max_depth: 8 | | | min_samples_leaf: 2 | | | min_samples_split: 2 | |
| MLP | Processed data PCA (n_attributes: 4) | initial_learning_rate: 0.01 | | | Solver: Adam | | | learning_rate_adjustment: Constant | | | hidden_layer_sizes: (12, 12, 12, 8) | | | activation_functions: relu | | | Alpha ( penalty): 0.01 | |
| ELM | Processed data Corr (n_attributes: 4) | n_neurons: (8) | | | activation_functions: relu | | | Alpha: 100 | |
| GBR | Processed data Corr (n_attributes: 4) | n_estimators: 250 | | | max_features: Sqrt | | | min_samples_leaf: 2 | | | max_depth: 2 | | | learning_rate: 0.2 | | | Loss: lad | |
| XGBR | Processed data Corr (n_attributes: 4) | cosample_bytree | | | Subsample | | | reg_lamda | | | reg_alpha | | | min_child_weight | | | learning_rate | |
| STACK 1 | Processed data Corr (n_attributes: 4) | 100 kernel: RBF | |
| STACK 2 | Processed data Corr (n_attributes: 4) | Alpha ( penalty): 0.001 | |
|
|