A Study On The Fusion Of Granular Computing Based On Rough Set And Neural Networks

Rough Set, as one of the Granular computing's three main models,have attracted much attention since its'born and has a broad application on machine learning, pattern recognition and other fields. The main contents include the approximation set, decision systems, data preprocessing and attribute reduction and so on. It is a effective method of dealing with incomplete, inaccurate data. Artificial Neural Network, belongs to the scope of soft computing the same as Rough set, advocate to exploit the tolerance for imprecision, uncertainty by devising methods of computation, which lead to an acceptable solution at low cost. With the Development over almost 70 years, Neural Network have been more mature, and widely used in many areas.The characteristics of Rough set theory and Artificial Neural Network are very bright and obviously complementary. Therefore, a certain way to implement the fusion of the two theories, using their advantages to make up for deficiencies, will be will be a a win-win results.This paper firstly introduces the basic components of granular computing based on rough set and neural network, including the core part of rough set theory-attribute reduction and several reduction methods, three major factors of neural networks, BP Neural Network's structure, learning rules and improvements.Then, on the basis of the content above, explores and demonstrates the feasibility of the fusion of rough set and neural networks; Enumerates the styles of the fusion and discusses three styles, serial connections, structural optimization, and rough neural networks in detail. Also, one of the structural optimization methods are made improvements, and proposes a structural optimization method.Finally, experimental verification. Experiment the fusion that using rough set as a preprocessor of neural networks through a instance of heart disease diagnosis. The network using train samples which have been treated by rough set attribute reduction diagnoses more accurately; Experiment the fusion that using rough set to optimize the neural network's structure. After optimization, the network's structure become simpler, and has faster processing speed. Both of the two experiments have proven the effectiveness of the fusion.