KDD procedures – KEX

 

KEX

Author(s):

Petr Berka, Milan Šimůnek

Responsibility:

Petr Berka (theory), Milan Šimůnek (software), ? (help)

Description:

KEX (Knowledge EXplorer) performs symbolic empirical multiple concept learning from examples, where the induced concept description is represented as a set of weighted decision rules. The created set of rules (rule base) can be used to classify unseen examples.

The input of procedure KEX consists of two parts:

  • a data matrix in a form of a database table,
  • a set of several parameters.

The KEX procedure generates automatically all the potentially usefull implications and tests each of them in the analyzed data to see if the impication can improve the set of decision rules learned so far. The resulting set of decision rules consists of several percent of all the generated implications.

The procedure KEX is realized by two modules:

  • The KEX-Task module creates the rule base and stores them in the metabase. It also tests the rules on training data.
  • The KEX-Result module reads the rule base and test it against testing data. This module is also used to classify a set of unseen examples. It is possible to sort the set of decision rules by various ways. It is also possible to copy results into clipboard and work with them in further ways.

The KEX differs from separate-and-conquer and divide-and-concquer machine learning algorithms in the fact, that covered examples are not removed from the data. That's why:

  • KEX can create more than one rule covering a specific example,
  • the knowledge base of KEX can contain both a rule and its subrule,
  • during consultation, the system can recommend (infer with weight > 0.5) more than one concept.

Because of used statistical test, KEX requires reasonable amount of input data.

Files to download:
LM.KEx.zip 1.96 MB April 21, 2017
KExTask and KExResult legacy modules for KnowedgeExplorer machine learning system

Print page

 
 

Send comments about this site to the webmaster