A new pattern recognition model based on heuristic SOM network and Rough Set Theory

Rough Set Theory is increasingly becoming one of the most effective tools for pattern recognition until now. While, in rough set theory, the discretization of continuous attributes is indispensable in any data preprocessing for the rough set computational intelligence algorithm and also critical to the quality of rule generated. To solve this problem, and based on the available researching achievements, we introduce the Self-Organizing feature Map (SOM) to cluster the original data into the required classes. In this paper, to direct the SOM clustering number of every continuous attribute, based on the indiscernibility discipline in Rough Set Theory, we define two new measurements-attribute Maximum Discernibility Value (MDV) and Attribute Redundancy Rate (ARR) as the strategic searching factros. MDV is employed to decide the heuristic strategy for the SOM neural network in the data preprocessing stage. And the ARR is for the attribute reduction as a effective feedback to the SOM clustering. Independent of domain experience, the combination of MDV, SOM, Skowron reduction, and the ARR can adjust the clustering number for every continuous attirbute automatically. Therefore, in theory, the computational speed is heightened greatly for the rough set attribute reduction. And in the end, a factual application case demonstrates the whole process effectively.