A rough neural expert system for medical diagnosis

Expert systems are the major practical application of artificial intelligence. In spite of the progress in expert system technology, the technology has some limitations in knowledge acquisition, inference, and level of intelligence, et al. In this paper, a rough neural expert system is constructed using rough set theory and neural networks. The methodology of rough set theory serves as a pre-processor for neural networks, including provision default values for missing data, discretization, binerization, attribute reduction and data transformation for network input. Knowledge acquisition is accomplished with the learning program of neural network. Then, the trained network serves as a knowledge base of the system. In the end, using a real example of diagnosis of coronary artery disease, a rough neural expert system is designed. The construction process of the system is illustrated in detail. The system correctly classified 83.75% of the testing set at a tolerance level of 0.25, and 85% at a tolerance level of 0.30.