How to Find Important Attributes with DBMS_PREDICTIVE_ANALYTICS

Before building a model, it helps to know which attributes have anything to do with the target. DBMS_PREDICTIVE_ANALYTICS.EXPLAIN answers that in one call, without a model: it ranks every column of a table or view by how much it explains the target column, and writes the ranking to a result table.

Code for This Guide

The main examples are in the examples/machine-learning folder of the Oracle Database 26ai code repository on GitHub, each with its output. They use NIMBUS, the sample schema of a fictional airline, which you can install with the scripts in the setup/nimbus folder.

They come from Oracle Database 26ai SQL and PL/SQL Book.

Syntax

dbms_predictive_analytics.explain(
  data_table_name     => 'TABLE_OR_VIEW',
  explain_column_name => 'TARGET',
  result_table_name   => 'RESULT_TABLE');
-- RESULT_TABLE: ATTRIBUTE_NAME, EXPLANATORY_VALUE, RANK

The result table must not exist yet. Explanatory values run from 0, no relation, to 1.

The example uses the ML_TICKETS view from How to Explain Predictions with PREDICTION_DETAILS in Oracle.

What Explains the Cabin?

Example:

create or replace view ml_cabin_factors as
  select cabin, tier, points, days_ahead, distance_km, legs, age from ml_tickets;

begin
  dbms_predictive_analytics.explain(
    data_table_name     => 'ML_CABIN_FACTORS',
    explain_column_name => 'CABIN',
    result_table_name   => 'CABIN_EXPLAINED');
end;
/
select attribute_name, round(explanatory_value, 4) as explanatory_value, rank
from   cabin_explained
order  by rank;

Output:

View ML_CABIN_FACTORS created.

PL/SQL procedure successfully completed.

ATTRIBUTE_NAME       EXPLANATORY_VALUE    RANK
_________________ ____________________ _______
DAYS_AHEAD                      0.3411       1
POINTS                          0.2219       2
TIER                            0.2163       3
LEGS                                 0       4
AGE                                  0       4
DISTANCE_KM                          0       4

6 rows selected.

Days booked ahead explains the cabin best, then loyalty points and tier. Legs, age, and distance explain nothing on their own. This matches what the decision tree found: its predictions rest on days ahead and points. The example drops the result table afterward.

Things to Know

• EXPLAIN looks at one attribute at a time, so attributes that matter only in combination can rank low.
• Use it to choose attributes before training, and to check that a model relies on sensible ones.
• Drop the ML_CABIN_FACTORS view when done.

Related Guides

• How to Explain Predictions with PREDICTION_DETAILS in Oracle
• How to Evaluate Models with COMPUTE_CONFUSION_MATRIX in Oracle

Conclusion

DBMS_PREDICTIVE_ANALYTICS.EXPLAIN ranks a table's columns by how much each explains a target, without building a model. Use it to pick attributes and to sanity-check what your models rely on.