Predicting WWW surfing using multiple evidence combination
The VLDB Journal — The International Journal on Very Large Data Bases
Testing terrorism theory with data mining
International Journal of Data Analysis Techniques and Strategies
Optimal top-k generation of attribute combinations based on ranked lists
SIGMOD '12 Proceedings of the 2012 ACM SIGMOD International Conference on Management of Data
| Hi-index | 0.00 |
In this work we apply Dempster-Shafer's theory ofevidence combination for mining medical data. Weconsider the classification task in two domains: Breasttumors and skin lesions. Classifier outputs are used as abasis for computing beliefs. Dynamic uncertaintyassessment is based on class differentiation. We combinethe beliefs of three classifiers: k-Nearest Neighbor(kNN), Naïve Bayesian and Decision Tree. Dempster'srule of combination combines three beliefs to arrive atone final decision. Our experiments with k-fold crossvalidation show that the nature of the data set has abigger impact on some classifiers than others and theclassification based on combined belief shows betteroverall accuracy than any individual classifier. Wecompare the performance of Dempster's combination(with differentiation-based uncertainty assignment) withthose of performance-based linear and majority votecombination models. We study the circumstances underwhich the evidence combination approach improvesclassification.