Local mean representation based classifier and its applications for data classification
| Autor: | Juliang Hua, Qian Chengshan, Geng Yang, Pu Huang |
|---|---|
| Rok vydání: | 2016 |
| Předmět: |
business.industry
Data classification Nonlinear dimensionality reduction 020206 networking & telecommunications Computational intelligence Linear classifier Pattern recognition 02 engineering and technology Quadratic classifier computer.software_genre Nonlinear system Artificial Intelligence 0202 electrical engineering electronic engineering information engineering 020201 artificial intelligence & image processing Computer Vision and Pattern Recognition Artificial intelligence Data mining Linear combination business Classifier (UML) computer Software Mathematics |
| Zdroj: | International Journal of Machine Learning and Cybernetics. 9:969-978 |
| ISSN: | 1868-808X 1868-8071 |
| DOI: | 10.1007/s13042-016-0621-0 |
| Popis: | Data classification is a fundamental problem in many research areas. This paper proposes a novel classifier, namely local mean representation based classifier (LMRC), for data classification. Based on the concept that neighboring samples should have most similar properties and for a testing sample, the similar properties should be concentrated on the mean of its intra-class nearest neighbors, LMRC represents the testing sample as a linear combination of its all local class means and assigns the testing sample to the class associating with the biggest item of the linear combination coefficient vector. LMRC is easy to employ with a least squares estimator, and it needs not to tune any parameter and could explore the local neighborhood relationship between samples to enhance the classification performance. Furthermore, to deal with the nonlinear problems, we extend the linear LMRC to its kernel version called kernel LMRC (KLMRC). Experiments on some benchmark datasets validate the superiority of the proposed two methods over other state-of-the-art methods. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|
Full text from SpringerLink