The Research And The Application Of The Fuzzy Support Vector Machines

Support vector machine (SVM), which was proposed by Vapnik and some other scholars,is one of the standard tools for machine learning and data mining. It is an implementation ofstructure risk minimization principle in the statistical learning theory. Based on the statisticallearning theory and optimization theory, SVM has been successfully applied to many fieldssuch as pattern recognition, regression and so on. Because of its excellent learning power, thistechnology has turned into the topic of machine learning, and also gained successfulapplications in many fields, such as handwriting digit recognition, voice recognition, facedetection, and etc.But as a new technology, SVM, which still has some limitations, needs to be furtherexplored and perfected. The standard Support Vector Machine methodology that learnsclassification problem is originally designed for binary classification, not directly applied to amulti-class classification. Although there are already some successful multi-class classificationmethods such as One-versus-One, One-versus-Rest, DDAGSVMs and so on, they are used toconstruct and combine a series of several binary classifiers by some way to realize a multi-classclassification. How to extend it to a multi-class classification effectively is a topic being studied.Moreover, there is lots of fuzzy information in the objective world. And if fuzzy information(fuzzy parameters) exists in the training set of SVMs, the traditional SVMs will fail. So theresearch of the fuzzy support vector machine and multi-class classification has greatsignificances.Much attention has been paid to the research and the application of the fuzzy supportvector machine to a multi-class. The paper puts forwards a multi-class classification based onfuzzy support vector machine, and applies fuzzy C-means clustering algorithm to fuzzifytraining samples. The results of experiment show that the new type of fuzzy support vectormachine in this dissertation is more effective. This new method is applied to evaluate thecreativity degree of architectural outline conceptual design.In conclusion, new contributions of this paper are listed as follows:1. Put forward a multi-class classification method based on fuzzy support vector machines.Fuzzy support vector machines are combined with binary tree architecture. Consequently,a modified Self-Organization Map (SOM), KSOM (kernel-based SOM), is introduced toconvert the multi-class problem into binary tress, in which the binary decisions are made byFSVMs. The results of the experiment show that the above method we propose is feasible andeffective.2. Fuzzify the samples by fuzzy C-means clustering algorithm.The fuzzy C-means clustering algorithm is introduced to fuzzify each train sample and testsample, and with the classification, we recalculate its membership degree to different classes.3. Evaluate the creativity degree of architectural outline conceptual design based oncreative conceptual design.As for the evaluation of the creative degree in conceptual design, the paper extracts thequalified characteristics of the architectural outline, rescales all characteristics with the specifiedrange, usually [-1,1], fuzzifies the characteristics as the train samples, and evaluates thecreativity degree of the conceptual design of the architectural outline by using multi-classclassification method based on fuzzy support vector machines. The experiment enunciates thatthe result of evaluation is accurate and the conclusion of this experiment is satisfactory.