Summary: | 碩士 === 中華大學 === 資訊管理學系(所) === 96 === In this thesis we proposed a multi-classifier fusion framework based on Dempster-Shafer theory. Using the training data, we computed the local district predication rates which used as certainty measures in later data fusion for each classifier. We verified the proposed method with the KDD Cup’99 data from Massachusetts Institute of Technology Lincoln Laboratory. By cooperating five feature selection methods (Principal Component Analysis, PCA; Multiple Linear Regression, MLR; Discriminant Analysis, DA; Rough Set Theory, RST; Genetic Algorithms, GA), five kinds of the examined feature sets were derived. Support Vector Machine (SVM) is the core classification tool. Finally, data fusion was implemented by Dempster-Shafer theory the method to integrate the classified results derived from the aforementioned classifiers. Our result proves that the data fusion using Dempster-Shafer theory with the certainty measures is effective. The prediction rate can be improved. The proposed method is better than voting method and probability averaging method.
|