Study On Knowledge Discovery And Its Applications Based On The Rough Set And Neural Network

Rough set theory, which was proposed by Pawlak in early 1980's, is a mathematical tool of dealing with uncertain and vague information. Knowledge reduction is the core of Rough set. It obtains the reduction attribute and classification rules while holding the ability of classification unchanged.Neural network obtains knowledge by simulating neuron and the established model is intelligent. The main learning process is adjusting the weight and bias of the neuron node. Neural network can be applied to restore the distributed information and process the parallel and cooperating information. It is self-organized and combines message processing and information storage.Rough set theory and Neural network are widely used. However, Rough set is sensitive with noise data, and Neural Network is overtrained when there are many redundant data in data set. This thesis combines the two methods' merit, studies Rough set theory and Neural network jointly.The main work of this thesis includes:(l)The reduction methods of knowledge are introduced, the method based on exclusive-OR matrix and the method based on information content are presented. The reduction method based on the branch bounding idea is proposed. It is implemented by using the UCI datas.(2)The learning and training method with BP Neural Network is introduced. The improved algorithm of BP Network is presented. The model of Neural Network is established by using the LM algorithm in Matlab 7.0.(3)This thesis combines Rough set theory and Neural network perfectly. Firstly, the reduction rule is gained by Rough set theory. Then, the rules are sent to Neural Network, and the model of Roughset-NN is established. The statistical data of poplar tree's growth is handled with this model, and the prediction of poplar tree's growth can be obtained with it.