Table 2 Main hyperparameters of the prediction models
From: A novel stacking ensemble learner for predicting residual strength of corroded pipelines
Model | Hyperparameters | Meta-learner |
|---|---|---|
KNN | Number of neighbors = 2, Weight = distance−1 | - |
SVR | Penalty factor = 969.38, Kernel = Radial Basis Function, Kernel width = 0.104 | - |
RF | Number of trees = 18, Feature ratio = 0.591, Maximum leaf nodes = 179, Random seed = 749,811 | - |
ETR | Number of trees = 6, Feature ratio = 0.449, Maximum leaf nodes = 180, Random seed = 468,154 | - |
MLP | Hidden layer sizes = (50,90,80), Initial learning rate = 0.006, Random seed = 23,652 | - |
LightGBM | Number of trees = 203, Number of leaves = 7, Minimum child samples = 3, Learning rate = 0.219, Feature sampling ratio = 0.694, L1 regularization = 0.034, L2 regularization = 0.074 | - |
XGBoost | Number of trees = 340, Max leaves = 13, Min child weight = 2.785, Learning rate = 0.323, Sample ratio = 0.658, Feature sampling ratio (level) = 1.0, Feature sampling ratio (tree) = 0.868, L1 regularization = 0.001, L2 regularization = 2.19 | - |
Stacking-KNN | Number of neighbors = 5, Weight = distance−1 | KNN |
Stacking-Ridge | Regularization strength = 45.155 | Ridge |
Stacking-XGBoost | Number of trees = 899, Maximum depth = 6, Learning rate = 0.02, Sample ratio = 0.525, Feature sampling ratio = 0.557 | XGBoost |
