Inducing decision rules: a granular computing approach

Abstract

Inducing rules is one of the key methods of discovering information hidden in data. In this paper, a method is proposed for inducing decision rules and decision algorithms by a granular computing approach, based on a decision logic language in information tables. And we prove that in consistent information tables, the induced decision algorithms are consistent and complete, and the decision algorithms induced by different partitions are equivalent. Secondly, this paper studies two specific kinds of partitions: partitions inducing atomic decision algorithms and partitions inducing the most general decision algorithms. An algorithm is given for finding the partitions inducing atomic decision algorithms which are also very close to the partitions inducing the most general decision algorithms. The partitions obtained using this algorithm can induce the decision rules which are all atomic, and whose number will be close to the lowest possible. This is then a solution to the problem of finding the simplest decision rules and algorithms.