This function is for use with classification models created using the
DBMS_DATA_MINING package or with the Oracle Data Mining Java API. It is not valid with other types of models. It returns a varray of objects containing all classes in a multiclass classification scenario. The object fields are named
COST. The datatype of the
PREDICTION field depends on the target value type used during the build of the model. The other two fields are both Oracle
NUMBER. The elements are returned in the order of best prediction to worst prediction.
bestN, specify a positive integer to restrict the returned target classes to the
N having the highest probability. If multiple classes are tied in the Nth value, the database still returns only
N values. If you want to filter only by
NULL for this parameter.
cutoff, specify a
NUMBER value to restrict the returned target classes to those with a cost less than or equal to the specified cost value. You can filter solely by
cutoff by specifying
When you specify values for both
cutoff, you restrict the returned predictions to only those that are the
bestN and have a probability (or cost when
MODEL is specified) surpassing the threshold.
MODEL to indicate that the scoring should be performed by taking into account the cost matrix that was associated with the model at build time. If no such cost matrix exists, then the database returns an error.
When you specify
cutoff are treated with respect to the prediction cost, not the prediction probability. That is,
bestN restricts the result to the target classes having the N best (lowest) costs, and
cutoff restricts the target classes to those with a cost less than or equal to the specified cutoff.
When you specify this clause, each object in the collection is a triplet of scalar values containing the prediction value (the datatype of which depends on the target value type used during model build), the prediction probability, and the prediction cost (both Oracle
If you omit
MODEL, each object in the varray is a pair of scalars containing the prediction value and prediction probability. The datatypes returned are as described in the preceding paragraph.
mining_attribute_clause behaves as described for the
PREDICTION function. Please refer to mining_attribute_clause.
The following example lists, for ten customers, the likelihood and cost of using or rejecting an affinity card. This example has a binary target, but such a query is also useful in multiclass classification such as Low, Med, and High.
This example and the prerequisite data mining operations can be found in the demo file
$ORACLE_HOME/rdbms/demo/dmdtdemo.sql. General information on data mining demo files is available in Oracle Data Mining Administrator's Guide. The example is presented here to illustrate the syntactic use of the function.
SELECT T.cust_id, S.prediction, S.probability, S.cost FROM (SELECT cust_id, PREDICTION_SET(dt_sh_clas_sample COST MODEL USING *) pset FROM mining_data_apply_v WHERE cust_id < 100011) T, TABLE(T.pset) S ORDER BY cust_id, S.prediction; CUST_ID PREDICTION PROBABILITY COST ---------- ---------- ----------- ----- 100001 0 .96682 .27 100001 1 .03318 .97 100002 0 .74038 2.08 100002 1 .25962 .74 100003 0 .90909 .73 100003 1 .09091 .91 100004 0 .90909 .73 100004 1 .09091 .91 100005 0 .27236 5.82 100005 1 .72764 .27 100006 0 1.00000 .00 100006 1 .00000 1.00 100007 0 .90909 .73 100007 1 .09091 .91 100008 0 .90909 .73 100008 1 .09091 .91 100009 0 .27236 5.82 100009 1 .72764 .27 100010 0 .80808 1.54 100010 1 .19192 .81 20 rows selected.