Research For Attribute Reduction Based On Relational Databases

Nowadays, followed by rapid development of Internet and transmission styles added, more and more information could rush to people rapidly. Mass and variety data bring more convenience to human, but at the same time, it brings us an "Information Bomb". So many people look forward to seeking a way to find the key part to instead of them. In this situation, Data Mining emerged and is used in more and more fields.Attribute reduction is one of the most important ways in Data Ming. A reduct is a set of attributes that preserve partition. It means that a reduct is a minimal subset of attributes which has the same classification ability as the whole set of attributes in universe. So it can use less data to present the same information we need.With the description of the important role of information reducing in modern time, this chapter analyses on the important role of core attributes, the cost on calculating core attributes by discernibility matrix and the positive area and gives a simple description of current classic algorithms. It puts forward a necessary and sufficient condition on whether a subset of attributes contains the core attributes. Base on these researches'results and referring the reality that most of the digital data is stored in database currently, an improved attribute reduction algorithm based on Rough set is presented, using the superiority of the technology on database. Moreover, we propose a new algorithm based on the unnecessity of the calculating on the core and the complement of the positive area.Experiments'results show that the two algorithms are more efficient in the larger or largest data sets. They resolve the dilemma of the current attribute reduction and less complex and can be easily realized in advantage of database query language. It will promote the development in the area of Data Ming, artificial intelligence, machine learning and other areas.