Table 4 Grid-search parameters for each candidate algorithm
Algorithms | Hyperparameter space to improve algorithm performance |
|---|---|
DT | Measure of impurity = ['gini', 'entropy'] & Split strategy = ['best', 'random'] & Max depth = [None, 3] & Max features = ['auto', 'sqrt', 'log2'] & Class weight = ['balanced', None] |
RF | Measure of impurity = ['gini', 'entropy'] & Bootstrap = [False, True] & Max depth = [None, 3] & Warm start = [False, True] & Class weight = ['balanced', 'balanced_subsample'] & Number of trees = [100, 200, 300, 400] |
ET | Measure of impurity = ['gini', 'entropy'] & Max depth = [None, 3] & Warm start = [False, True] & Class weight = ['balanced', 'balanced_subsample'] & Number of trees = [100, 200, 300, 400] |
LR | Optimization algorithm = ['newton-cg', 'lbfgs', 'liblinear', 'sag', 'saga'] & Inverse of regularization strength = [0.1, 1, 10, 100] & Class weight = ['balanced', None] & Dual = [False, True] & Fit intercept = [False, True] & Tol = [0.001, 0.01] & Warm start = [False, True] |
GB | Measure of impurity = ['friedman_mse', 'mse', 'mae'] & Max depth = [None, 3] & Number of trees = [100, 200, 300] |
ANN | Alpha = [0.1, 0.01, 0.001] & Activation = [‘relu’, ‘logistic’] & Early stopping = [True, False] & Number of hidden layers = [200, 300, 400] |
QDA | *No parameter settings available in Scikit-learn implementation |