Experiments in Polychotomous Classification
نویسندگان
چکیده
A central problem in pattern recognition is the classi cation of a feature vector into one of several possible classes. In polychotomous classi cation the number of possible classes is higher than two. Given a set of training vectors with known classes, a classi er can be constructed using for example some density estimation or regression method. In a typical approach one estimates, using all training data from all classes, the posterior probability functions of each class and then classi es future feature vectors according to the highest estimated posterior probability. In some recent proposals a di erent approach is suggested where one rst estimates classi ers only for each pair of classes and then combines these into a nal polychotomous classi er (Friedman 1996, Hastie and Tibshirani 1998). The advantages of the pairwise approach are hoped to come from both increased classi cation accuracy and reduction in computational complexity. We report experimental results using various classi er designs and a variety of data sets arising in di erent applications, including automatic recognition of handwritten digits (Holmstrom et al. 1997).
منابع مشابه
The Analysis of Bayesian Probit Regression of Binary and Polychotomous Response Data
The goal of this study is to introduce a statistical method regarding the analysis of specific latent data for regression analysis of the discrete data and to build a relation between a probit regression model (related to the discrete response) and normal linear regression model (related to the latent data of continuous response). This method provides precise inferences on binary and multinomia...
متن کاملPairwise classifier combination using belief functions
In the so-called pairwise approach to polychotomous classification, a multi-class problem is solved by combining classifiers trained to discriminate between each pair of classes. In this paper, this approach is revisited in the framework of the Dempster-Shafer theory of belief functions, a non-probabilistic framework for quantifying and manipulating partial knowledge. It is proposed to interpre...
متن کاملA Phase I Dose-_finding Study Based on Polychotomous Toxicity Responses A Phase I Dose-finding Study Based on Polychotomous Toxicity Responses
متن کامل
Classification by Pairwise Coupling
We discuss a strategy for polychotomous classification that involves estimating class probabilities for each pair of classes, and then coupling the estimates together. The coupling model is similar to the Bradley-Terry method for paired comparisons. We study the nature of the class probability estimates that arise, and examine the performance of the procedure in simulated datasets. The classifi...
متن کاملA Multi-scale Nonparametric/Parametric Hybrid Recognition Strategy with Multi-category Posterior Probability Estimation
The synthesis of an effective multi-category nonlinear classifier with the capability to output calibrated posterior probabilities to enable post-processing is of great significance in practical recognition situations in that the posterior probability reflects the assessment uncertainty. In this paper, a multi-scale nonparametric and parametric hybrid recognition strategy is developed for this ...
متن کامل