Diversity-based Feature Partitioning for Combination of Nearest Mean Classifier

Husin, Abdullah (2013) Diversity-based Feature Partitioning for Combination of Nearest Mean Classifier. Information Engineering Letters, 3 (3). ISSN 2163-4114

Text icet.pdf Download (13MB)

Abstract

Nearest Mean Classifier (NMC) provides good performance for small sample training. However concatenate different features into a high dimensional feature vectors and process them using a single NMC generally does not give good results because dimensionality problem. Most methods used to address the dimensionality problem focuses on feature selection method, choosing a single feature subset, while ignoring the rest. Although there are several algorithms have been proposed, but there are drawbacks to using of feature selection method. The assumption that a large set of input features can be reduced to a small subset of relevant features is not always true. In some cases the target feature is actually affected by most of the input features and removing features will cause a significant loss of important information. Thus, the classifier may achieve a lower level of accuracy than the classifiers that accesses all the relevant features. In this method the feature set clusters into different feature subset. NMC ensembles constructed by assigning each individual classifier in the ensemble with a cluster of different feature subset. The advantage of this approach is that all of available information in the training set is used. There is no irrelevant feature in the training set are eliminated. Based on experimental results the new technique significantly improve the nearest mean classifier (NMC) with 95% confidence.

Actions (login required)

View Item