The function calculates the values of prob_response and predict_confidence with the following formulas.
This is the formula for the value of class r:
valuer = Wr·X
where:
- X is the vector of predictor values corresponding to an observation.
- Wr is the vector of predictor weights calculated by the model for class r, where r is a class specified by the Responses syntax element.
To see the predictor weights, use the function SVMSparseSummary (ML Engine) or SVMDenseSummary (ML Engine), with OutputSummary ('false').
prob_response
For binary classification, the formula for the probability that a response belongs to class r is:
prob_response = sigmoid (valuer)
For multiple-class classification, the formula for the probability that a response belongs to class r is:
prob_response = softmaxr (values)
predict_confidence
The column predict_confidence, which appears only if you omit the Responses syntax element, displays the probability that the observation belongs to the class in the column predict_value. This value is the maximum value of prob_response over all responses r.