Data mining as the tool to extract useful information from a large amount of data is being an important issue nowadays. Particularly, classification is an important theme in data mining. In classification, rough sets theory has provided mathematical t...
Data mining as the tool to extract useful information from a large amount of data is being an important issue nowadays. Particularly, classification is an important theme in data mining. In classification, rough sets theory has provided mathematical tools to deal with uncertainty and vagueness in large data set.
The problem to find minimal subsets (called "reduct") of attributes that can describe all of the concepts in the given data set is known as NP-hard problem. However, algorithms to reduce the computation intensity have been proposed. A Binary Discernibility Matrix(BDM) method is one of these algorithms.
The original BDM method doesn't consider the correlation and association between attributes and classes. The execution order of attributes is decided only by the discernibility degree. However, in cases in which attributes have bised values, the discernibility degree value becomes low. As a result, they are excluded in the reduct. These are not also included in the set of classification rules.
This paper proposes a classification algorithm using an improved Binary Discernibility Matrix(BDM) method. In this proposed method, we consider the confidence and lift to find attributes having the highest association, which are then executed first. Therefore, these attributes are included in the reduct because they are indispensable attributes. We make up for the shortcomings of the original BDM method.
Classification rules should be explicit and understandable. The experiment comparing the original BDM with the improved BDM shows that the proposed method has better performance in accuracy and it creates simple and clear classification rules.