SVMACHINE Procedure

This example shows how you can use PROC SVMACHINE to create scoring code that can be used to score future home equity loan applications. The data set Hmeq, which is in the Sampsio library that SAS provides, contains observations for 5,960 mortgage applicants. A variable named Bad indicates whether the customer has paid on the loan or has defaulted on it. Table 5 describes the variables in Hmeq.

You can load the Hmeq data set into your CAS session by specifying your CAS engine libref in the second statement in the following DATA step:

data mylib.hmeq;
   set sampsio.hmeq;
run;

The following statements execute the SVM algorithm on the mylib.hmeq data table:

filename codefile temp;
proc svmachine data=mylib.hmeq;
   input reason job derog delinq ninq / level=nominal;
   input loan mortdue value yoj clage clno debtinc / level=interval;
   target bad / desc;
   code file=codefile;
run;

The first INPUT statement defines the input variables Reason, Job, Derog, Delinq, and Ninq as categorical variables. The second INPUT statement defines the input variables Loan, MortDue, Value, YoJ, CLAge, CLNo, and DebtInc as continuous variables. The TARGET statement defines Bad (which is a binary variable) as the target variable and specifies the order of the target variable as descending. The CODE statement generates DATA step scoring code and stores it in the filename codefile. The scoring code can be used to score other home equity loan applications.

PROC SVMACHINE generates several ODS tables, some of which are shown in Output 39.1.1 through Output 39.1.5.

The "Model Information" table in Output 39.1.1 shows that the kernel function is linear, and the penalty parameter value is 1 (both of which are default values).

The observations table in Output 39.1.2 shows that the total number of observations is 5,960 and the number of observations used in the training is 3,364.

The "Training Results" table in Output 39.1.3 shows the inner product of weights, bias, total slack, and so on.

The "Misclassification Matrix" table in Output 39.1.4 displays the original observations and predicted values. Here the true positive is 43, the false negative is 257, the true negative is 3,055, and the false positive is 9.

The "Fit Statistics" table in Output 39.1.5 shows information about the accuracy, error, sensitivity, and specificity.

In addition to these ODS tables, PROC SVMACHINE also generates the "Optimization Iteration History" table. The CODE statement in this example generates SAS code and stores at in the filename codefile. Advanced SAS users can use the SAS code to easily score their data.