Perturbation-based classifier
- PDF / 546,568 Bytes
- 12 Pages / 595.276 x 790.866 pts Page_size
- 87 Downloads / 186 Views
METHODOLOGIES AND APPLICATION
Perturbation-based classifier Edson L. Araújo1,2 · George D. C. Cavalcanti1 · Tsang Ing Ren1
© Springer-Verlag GmbH Germany, part of Springer Nature 2020
Abstract The Bayes classifier depends on the conditional densities and the prior probabilities. Among many density functions, the Gaussian density has received more attention mainly motivated by its analytical tractability. The parameters of the Bayes classifier for the Gaussian distribution data are generally unknown, and approximations are calculated for the mean vector ˆ When a pattern is inserted in the training set of the class ωi , the values of the parameters ˆ and the covariance matrix Σ. µ ˆ i and ΔΣˆ i , respectively. The insertion of one pattern can cause a perturbation, ˆ i and Σˆ i change by an amount given by Δµ µ so we claim that this perturbation can be used for supervised classification purposes. Based on this assumption, we propose a supervised classifier called Perturbation-based Classifier PerC that assigns the class of the query pattern as the one that presents the smallest perturbation among all the classes after the insertion of this query pattern in the classes. The rationale is that the addition of a pattern that belongs to one specific class should not alter much the distribution of that class. PerC ˆ i and ΔΣˆ i ) to evaluate the class of a query pattern; so, it is a parameter-free classifier. The only uses the perturbations (Δµ proposed method was assessed on 21 datasets from the UCI Machine Learning Repository, and its results were compared with classifiers from the literature. Results have shown that PerC obtains very competitive recognition rates. Keywords Pattern recognition · Perturbation · Normal distribution · Bayes theory
1 Introduction Pattern recognition consists of two fundamental tasks: description and classification. Suppose an object under analysis, a pattern recognition system produces a description of it, the pattern description phase, and then classifies it according to that description, the recognition phase. The essential problem in pattern recognition is identifying an object as a member of a particular group, assuming that objects from the same group have more attributes in common than any object from other groups (Duda et al. 2012; Fukunaga 1972). Communicated by V. Loia.
B
George D. C. Cavalcanti [email protected] Edson L. Araújo [email protected] Tsang Ing Ren [email protected]
1
Centro de Informática, Universidade Federal de Pernambuco, Av. Jornalista Anibal Fernandes s/n, Recife, Brazil
2
Universidade Federal do Vale do São Francisco, Av. Antonio Carlos Magalhães, 510, Juazeiro, Brazil
Among various approaches for the recognition phase, the ones using statistical assumptions have been extensively studied and applied (Jain et al. 2000; Evgeniou et al. 2002). For instance, other approaches such as artificial neural networks (de Jesus Rubio 2017; Ding et al. 2017; Kumar et al. 2017), bayesian networks (Cooper and Herskovits 1992; Friedman et al. 1997; Perez et al. 20
Data Loading...
