Study Of Reduction Algorithms Based On Discernibility Matrix Of Length Constraint

With the coming of information age, a information world has appeared around people and the data of informaion system increase like exploding.How to analyse and mine the vast data efficiently is the main subject of current data mining field. More and more attention has been paid to the rough set theory in data mining field which proposed by professor Pawlak.Z, and some good application has appeared.The rough set theory is a new mathematics tool which is used to process fuzzy and indetermination problem. This theory which's advantages lie in not requiring prior information when carries out the classicification is to derive classification rules of conception by knowledge reduction without changing the classification capacity of the information system. The main contribution of this dissertation is to research the problem of attributes reduction in rough set theory and gives an improved algorithm by analysing the normal attributes reduction algorithm. There are the detailed contributions of this dissertation as follows:(1) The current situation and development, existing problems and development tendency of rough set theory in future has been studied and described, the basic rough set theory is also introduced.(2) The classical reduction algorithm based on discernibility matrix, the reduction algorithm based on information entropy and the reduction algorithm based on attributes frequency has been studied in this dissertation. Then, A improved reduction algorithm based on attributes frequency is given.Finally,This dissertation presents a reduction algorithm based on attribute buckets.(3) A reduction algorithm based on discernibility matrix of length constraint (RABDMLC) is presented. It is demonstrated that RABDMLC is effective through the contrastive experiment and the analyse application in the network fault alarm system .