Abstract: The concept lattice is accurate and complete in the knowledge representation, and it is an effective tool for data analysis and knowledge discovery. Classification rule mining based on the concept lattice in an efficient way is a challenging task. At present, there are many extraction algorithms for classification rule based on the concept lattice, but the number and form of extracted rule can not achieve satisfactory results. This paper focused on classification rule mining using intent reduction of the formal concept. It proposed an incremental way to compute intent reduction by adding object to the formal context one by one. Through modifying the incremental computation of the intent reduction of concepts, new algorithms are developed to compute the intent reduction, and then definitions of the exact classification rule base and the approximate base are given. On the basis, two algorithms for the exact and the approximate mining bases are designed. In the algorithms, only those concepts that concern the classification rule mining need to be considered in compute their intent reduction. Furthermore, the use of classification rule bases reduces the total number of classification rules that need to be extracted. In order to verify the data mining methods of classification association rules which were put forward in this research, the algorithms mentioned above were implemented by using C++. Finally, empirical experiments on UCI data demonstrated the efficiency of these algorithms in the mining of affirmative classification rules and approximate classification rules using intent reduction.
|
Received: 21 July 2009
|
Corresponding Authors:
handsome
|
|
|
|