Table 3 Hyper-parameters of different models in CP-EGBM optimized by mPSO for all the datasets.
From: An efficient churn prediction model using gradient boosting machine and metaheuristic optimization
Model | Function | DS 1 | DS 2 | DS 3 | DS 4 | DS 5 | DS 6 | DS 7 |
|---|---|---|---|---|---|---|---|---|
SVMRBF | Regularization (\(\mathrm{C}\)) | 100 | 156 | 50 | 65 | 25 | 120 | 87 |
Kernel coefficient (\(\gamma\)) | 0.213 | 0.302 | 0.030 | 0.001 | 0.003 | 0.203 | 0.137 | |
GBM | Number of estimators | 315 | 503 | 223 | 418 | 438 | 250 | 305 |
Learning rate | 0.093 | 0.103 | 0.132 | 0.312 | 0.034 | 0.001 | 0.003 | |
Max. depth of DTs | 5 | 5 | 4 | 6 | 6 | 7 | 6 | |
Min. samples for split | 5 | 8 | 6 | 10 | 7 | 8 | 9 | |
Max. features | 8 | 12 | 25 | 40 | 8 | 8 | 6 | |
Sub-sample | 1 | 0.82 | 0.90 | 0.95 | 0.83 | 0.97 | 0.83 |
