Attribute Reduction Based On Granular Computing Algorithm And Applied Research

Granular computing (GrC) has developed very rapidly in recent years, and many experts focus on its study. GrC makes it easy to deal with uncertain, fuzzy and incomplete informations, and it incorporates many other theories, such as artificial intelligence, fuzzy sets, rough sets et ceteral. GrC extracts general theory and basic elements from the problem and practice. The basic idea of granular computing mainly comes from the human habits of transfering big and difficult problems into many small and easy ones for solving. GrC is a structural, systematic approach to deal with complex problems, and it solves complex problems with a multi-angled and multi-level grain structure, so that it can give an approximately systematic solution for problems, which is very similar to what we often observe problem from different angles and levels.This thesis focused on the attributes reduction based on GrC. It is well known that the attribute reduction is a NP-hard problem. Here we used the granular computing combined with support vector machines to get a comparable result for attribute reduction of a system. Then we used the algorithm to the diagnosis of erythemato-squamous diseases. The works we did are here.First, we made an analysis of the current research and the background on attribute reduction and on granular computing and on the dignosis for erythema-squamous diseases. Then the basic concepts and basic theory of GrC and rough set theory were introduced in the thesis, as well as support vector machine (SVM), which we used as a classification tool in our study.Second, we analyzed the available attribute reduction methods, and proposed a new attribute reduction algorithm based on granular computing in this thesis. Our new algorithm used the difference between each condition attribute and decision attribute to measure the importance of the corresponding condition attribute, and ranked the condition attributes in descending order according to their importance. The sequential forward search strategy can be used to select the important attributes and remove the redundant ones from condition attributes to get the reduction of attributes. The algorithm stops as soon as the threshold is satisfied. We tested our algorithm using the datasets from UCI machine learning repository. SVM was used as a classification tool in our experiments. The experiment results proved that our new algorithm can get the best subset of attributes, and the decision information system can get higher classification accuracy with the attribute reduction we obtained via our new algorithm.Third, we still used the difference between each condition attribute and decision attribute to measure the importance of the corresponding condition attribute, and ranked the condition attributes in descending order according to their importance. The sequential forward search strategy can be used to select the important attributes, whilst the support vector machines were used to evaluate the cunrrent selelcted subset of attributes. Finally we choose the attribute subset that has the best classification performance as the optimal subset of attributs. Our algorithm combined the granular computing with support vector machines was applied to the differential diagnosis of erythemato-squamous diseases, where SVM was used as a classification tool. In our experiment we adopted a grid search technique to find the best parameters for SVM. The classification accuracy we got is up to 99.32% with only 23 attributes from the 34 features of the original dataset. It can be seen that the algorithm we proposed is efficient for the diagnosis of erythemato-squamous diseases.