Dempster-shafer Theory based Multi-class Support Vector Machines and their Applications

Authors:	Zhonghui Hu, Shanghai Jiaotong University, China Rupo Yin, Shanghai Jiaotong University, China Yuangui Li, Shanghai Jiaotong University, China Xiaoming Xu, Shanghai Jiaotong University, China
Topic:	6.4 Safeprocess
Session:	Signal Based Fault Detection and Isolation
Keywords:	Machine learning; Classifiers; Classification; Fault diagnosis; Diesel engines

Abstract

How to extend standard support vector machines to solve multi-classclassification problem and yield the outputs in the frame of Dempster-Shafer theory isuseful. The multi-class probability support vector machine is proposed, firstly. TheDempster-Shafer theory based multi-class support vector machine is designed byconstructing probability support vector machines for binary classification using oneagainst-all strategy and then combining them using Dempster-Shafer theory. Ourproposed method is applied to fault diagnosis for a diesel engine. The experimentalresults show our proposed method obtains a comparable performance with that ofstandard multi-class support vector machines. Furthermore, the uncertainty can also beevaluated.