Incremental learning of decision rules based on rough set theory

With the changes of databases, the former rule sets obtained from the data set require updating. We want that the algorithms of rule generation are incremental learning methodology for modification of the existing decision rules and their numerical measures when new objects are appended to the database, instead of running the whole learning process again. In this paper, based on rough set theory, the concept of partial derivative-indiscernibility relation is put forward in order to transform an inconsistent decision table to one that is consistent, called partial derivative-decision table, as an initial preprocessing step. Then the partial derivative-decision matrix is constructed. On-the basis of this, by means of decision function, an algorithm for incremental learning of rules is presented. The algorithm can also incrementally modify some numerical measures of a rule.