USE OF THE UNPIVOT RELATIONAL OPERATOR IN THE EFFICIENT GATHERING OF SUFFICIENT STATISTICS FOR DATA MINING

The invention concerns a method and apparatus for generating a tabulation of counts of occurrences of value combinations of a set of attributes over a relation consisting of a set of database records. The gathered counts (also referred to as sufficient statistics) of attribute occurrences or correlation counts is most preferably used in building a classification or density estimation model from the database records that can be used to predict some attribute values based on other attribute values. A new SQL operator designated the `UNPIVOT` operator operates by scanning the database records and for each record reorganizes that data to form an UNPIVOTED data record that include the combinations of attribute name, attribute value and the values for one or more selected class attributes. The UNPIVOTED table can be used to produce the desired sufficient statistics in one scan of the data using standard database engines. While materialization of UNPIVOTED table would cause a large added scan cost overhead, the UNPIVOT operator allows us to achieve the counts without the added cost by combining the UNPIVOT operator with other SQL `select` and `group by` operators the UNPIVOTED table can be counted without the need for materializing it. The result is a guaranteed one pass algorithm that does not incur the added scan cost factor. The savings in scan cost can extend to several orders of magnitude compared to other methodologies for getting the counts supported by current database engines. The sufficient statistics so gathered can be used to drive a variety of data mining algorithms.