Model Comparison Report
In the Support Vector Machines platform, the Model Comparison report contains the following information:
Show
A check box that indicates whether the Support Vector Machine Model report for the corresponding fit should appear in the report window.
Method
The model number.
Kernel Function
The kernel function for the corresponding model.
Cost
The value of the cost parameter for the corresponding model.
Gamma
The value of the gamma parameter for the corresponding model. If the model uses a linear kernel, this value is missing.
# SV
The number of support vectors used in the corresponding model.
Training Misclassification Rate
(Appears only if the response is categorical.) The misclassification rate for the observations in the training set. This rate is based on the classification decision rule that is calculated by the platform.
Validation Misclassification Rate
(Appears only if the response is categorical and validation is used.) The misclassification rate for the observations in the validation set. This rate is based on the classification decision rule that is calculated by the platform.
Test Misclassification Rate
(Appears only if the response is categorical and a test set is specified using a validation column.) The misclassification rate for the observations in the test set. This rate is based on the classification decision rule that is calculated by the platform.
Training RASE
(Appears only if the response is continuous.) The square root of the mean squared prediction error (Root Average Square Error) for the observations in the training set. See RASE.
Validation RASE
(Appears only if the response is continuous and validation is used.) The square root of the mean squared prediction error (Root Average Square Error) for the observations in the validation set. See RASE.
Test RASE
(Appears only if the response is continuous and a test set is specified using a validation column.) The square root of the mean squared prediction error (Root Average Square Error) for the observations in the test set. See RASE.
Validation Generalized RSquare
(Appears only if validation is used.) The Generalized RSquare value for observations in the validation set. See Fit Details.
Probability Threshold
(Appears only if the response is binary and validation is used.) The cutoff probability determined by the classification decision rule that is calculated by the platform. An observation is classified into the target level when its predicted probability exceeds this value.
Tip: You can change this value by using the Set Probability Threshold option in the Confusion Matrix report. See Confusion Matrix.
Conditional Validation Misclassification Rate
(Appears only if the response is binary and validation is used.) The misclassification rate for the observations in the validation set, conditioned on the probability threshold value.
Best
Indicates which model fit has the smallest misclassification rate. If validation is used without a test set, this is the model with the smallest validation misclassification rate. If validation is used with a test set, this is the model with the smallest test misclassification rate.
Model Comparison Plot
Shows a contour plot of the model performance of the validation set over the grid of parameter values. This report is available only if Tuning Design is specified in the Model Launch Control Panel.