Eigenclassifiers for combining correlated classifiers |
| |
| Authors: | Ayd?n Ula? Olcay Taner Y?ld?z Ethem Alpayd?n |
| |
| Affiliation: | 1. Department of Computer Engineering, Bo?aziçi University, TR-34342 Bebek, Istanbul, Turkey;2. Department of Computer Engineering, I??k University, TR-34980 ?ile, Istanbul, Turkey;1. Key Laboratory of Separation Science for Analytical Chemistry, Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Dalian 116023, China;2. School of Pharmaceutical Science and Technology, State Key Laboratory of Fine Chemicals, Dalian University of Technology, Dalian 116024, China;3. University of Chinese Academy of Sciences, Beijing 100049, China;1. School of Computer Science & Technology, Dalian University of Technology, 116024 Dalian, China;2. Key Laboratory of Separation Science for Analytical Chemistry, Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Dalian 116023, China;1. School of Information Science and Technology, Zhanjiang Normal University, Zhanjiang 524048, China;2. State Key Laboratory of Software Engineering, Wuhan University, Wuhan 430079, China;3. School of Mathematics and Computation Science, Zhanjiang Normal University, Zhanjiang 524048, China;4. School of Remote Sensing and Information Engineering, Wuhan University, Wuhan 430079, China;1. Department of Community Health Sciences, University of Calgary, Calgary, Alberta, Canada;2. Department of Family and Community Medicine, University of Toronto, Toronto, Ontario, Canada;3. Institute for Clinical Evaluative Sciences (ICES), Toronto, Ontario, Canada;4. Department of Family Medicine, McGill University, Montreal, Québec, Canada;5. Department of Medicine, University of Calgary, Calgary, Alberta, Canada;6. Department of Pharmacology and Therapeutics, University of Calgary, Calgary, Alberta, Canada;7. Department of Clinical Neurosciences, Hotchkiss Brain Institute, University of Calgary, Calgary, Alberta, Canada;8. Department of Medicine, University of Ottawa, Ottawa, Ontario, Canada;9. Division of General Internal Medicine, University of British Columbia, British Columbia, Vancouver, Canada;10. School of Public Health, University of Saskatchewan, Saskatoon, Saskatchewan, Canada;11. Manitoba Centre for Health Policy, University of Manitoba, Winnipeg, Manitoba, Canada;12. Surveillance and Assessment Branch, Alberta Health, Edmonton, Alberta, Canada;13. School of Public Health, University of Alberta, Edmonton, Alberta, Canada;14. Division of General Internal Medicine, University of Alberta, Edmonton, Alberta, Canada;1. Department of Micro- and Nanosciences, Aalto University, Tietotie 3, 02150 Espoo, Finland;2. Fraunhofer Institute for Solar Energy Systems ISE, Heidenhofstraße 2, 79110 Freiburg im Breisgau, Germany;3. Department of Engineering and Physics, Karlstad University, Universitetsg. 2, 65188 Karlstad, Sweden;1. Institute of Basic Research in Clinical Medicine, China Academy of Chinese Medical Sciences, Beijing 100700, China;2. Institute of Drug Research, Heilongjiang ZBD pharmaceutical Co. Ltd., Haerbin 150060, China |
| |
| Abstract: | In practice, classifiers in an ensemble are not independent. This paper is the continuation of our previous work on ensemble subset selection A. Ula?, M. Semerci, O.T. Y?ld?z, E. Alpayd?n, Incremental construction of classifier and discriminant ensembles, Information Sciences, 179 (9) (2009) 1298–1318] and has two parts: first, we investigate the effect of four factors on correlation: (i) algorithms used for training, (ii) hyperparameters of the algorithms, (iii) resampled training sets, (iv) input feature subsets. Simulations using 14 classifiers on 38 data sets indicate that hyperparameters and overlapping training sets have higher effect on positive correlation than features and algorithms. Second, we propose postprocessing before fusing using principal component analysis (PCA) to form uncorrelated eigenclassifiers from a set of correlated experts. Combining the information from all classifiers may be better than subset selection where some base classifiers are pruned before combination, because using all allows redundancy. |
| |
| Keywords: | |
| 本文献已被 ScienceDirect 等数据库收录！ |
|
