Table 2 Optimum hyper-parameter settings for experiments
From: Maintaining proper health records improves machine learning predictions for novel 2019-nCoV
Setting | AdaBoost | Bagging | Extra-Trees | Decision Tree | k-NN |
|---|---|---|---|---|---|
Base Estimator | None | None | NA | NA | NA |
# Estimators | 100 | 10 | 100 | NA | NA |
Learning rate | 2 | NA | NA | NA | NA |
Algorithm | SAMME.R | Bagging | Gini | Gini | KDTree |
Metric | Mean label accuracy | Mean label accuracy | Gini Impurity | Gini Impurity | Euclidean distance |
Random state | None | Random generation | None | Random generation | NA |
Max. samples to train needed to train base estimator | NA | 1 | NA | NA | NA |
Out-of-bag samples to estimate generalization error | NA | None | None | NA | NA |
Use whole ensemble to fit | NA | Yes | Yes | NA | NA |
# Jobs to run in parallel | NA | 1 | 1 | NA | 1 |
Random resampling | NA | 3141 | 12 | NA | NA |
Min. sample to be a leaf | NA | NA | 2 | 2 | NA |
Sample weighting | NA | NA | All equal, weight of 1 | All equal, weight of 1 | NA |
# of features for best split | NA | NA | Square root of the # of features | Max. features = # of features | NA |
Min. number of leaf nodes | NA | NA | Unlimited | NA | NA |
Split criteria | NA | NA | Impurity level > 0 | NA | NA |
Reuse previous call to fit and add more estimators to ensemble | NA | No | Yes | NA | NA |
Number of neighbours | NA | NA | NA | NA | 1 |