Stepwise logistic regression | Random forest | Neural network* | Elastic Net | ||||
---|---|---|---|---|---|---|---|
Software | SAS Enterprise Guide 7.1 Proc HPlogistic | SAS Enterprise Guide 7.1 Proc HPforest | SAS Enterprise Guide 7.1 proc HPNeural | R (caret package) | |||
Select criterion | Significance level | Max trees | 100 | Type | Fully connected feed forward | Alpha | 0–1 in steps of 0.1 |
Stop criterion | Significance level | Mas depth | 30 | Number of hidden layers | 1 | Lambda | 0.001 to 100.000 in logarithmic steps |
Effect hierarchy enforced | None | Prune threshold | 0.1 | Number of hidden neurons | 10–15 | Folds for crossvalidation | 10 |
Entry significance level (SLE) | 0.05 | Leaf fraction | 0.00001 | Number of weights | 7721 | Link function | Binomial |
Stay significance level (SLS) | 0.05 | Category bins | 30 | Optimization technique | Limited memory BFGS | ||
Stop horizon | 1 | Interval bins | 100 | Maxiter | 1000 | ||
Minimum category size | 5 | Activation function | Identity | ||||
Rows of sequence to skip | 5 | ||||||
Split criterion | Gini | ||||||
Preselection method | Loh |