Snippet Name: PREDICTION_COST
Description: PREDICTION_COST is for use with any classification model created by the DBMS_DATA_MINING package or with the Oracle Data Mining Java API. It returns a measure of cost for a given prediction as an Oracle NUMBER.
If you specify the optional class parameter, then the function returns the cost for the specified class. If you omit the class parameter, then the function returns the cost associated with the best prediction. You can use this form in conjunction with the PREDICTION function to obtain the best pair of prediction value and cost.
The COST clause is relevant for all classification models.
Specify COST MODEL to indicate that the scoring should be performed by taking into account the scoring cost matrix associated with the model. If no such scoring cost matrix exists, then the database returns an error.
Specify COST MODEL AUTO if the existence of a cost matrix is unknown. In this case the function returns the cost using the stored cost matrix if it exists. If no stored cost matrix exists, then the function applies the unit cost matrix (0's on the diagonal and 1's everywhere else). This is equivalent to one minus probability for the given class..
Use the VALUES clause (the bottom branch of the cost_matrix_clause) to specify an inline cost matrix. You can use an inline cost matrix regardless of whether the model has an associated scoring cost matrix.
Also see: » PREDICTION_SET
» PREDICTION_PROBABILITY
» PREDICTION_DETAILS
» PREDICTION_BOUNDS
» PREDICTION
» FEATURE_SET
» FEATURE_ID
» CLUSTER_SET
» CLUSTER_PROBABILITY
» CLUSTER_ID
Comment: (none)
Language: PL/SQL
Highlight Mode: PLSQL
Last Modified: March 07th, 2009

PREDICTION_COST(<schema.model>, <class> <cost_matrix_clause>
<mining_attribute_clause>)
WITH
cust_italy AS (
SELECT cust_id
FROM mining_data_apply_v
WHERE country_name = 'Italy'
ORDER BY PREDICTION_COST(DT_SH_Clas_sample, 1 COST MODEL USING *) ASC, 1
)
SELECT cust_id
FROM cust_italy
WHERE ROWNUM < 11;
CUST_ID

100081
100179
100185
100324
100344
100554
100662
100733
101250
101306
10 rows selected.
