Multiple Classifiers Approach based on DynamicSelection to Maximize Classification Performance

Abstract

In the past decade, many researchers have employed various methodologies to combine decisions of multiple classifiers in order to achieve high pattern recognition performance. However, two main strategies of combination are possible. The first strategy uses the different opinions of classifiers to make the final decision; it corresponds to classifiers fusion. The second strategy uses the decisions of one or more better classifiers in a specific region of feature space; it corresponds to the selection of classifiers. In this paper, we propose a dynamic multiple classifiers selection system organized in two levels of decision. Two classification methods are used: Semi-Supervised Fuzzy Pattern Matching (SSFPM) and Support Vector Machines (SVM). SSFPM is used to determine the ambiguous regions. Then, the patterns located in these regions are classified by SVM. The detection of the occurrence of new classes and the learning of their membership functions are achieved online using SSFPM. This combination helps to overcome the drawbacks of the both methods by gathering their advantages leading to increase the classification performance.

Comprehensive Study and Evaluation of Commonly used Dimensionality Reduction Techniques in Biometrics Field

“…LDA transforms a given dataset into a lower dimensional space with more advantage than the PCA technique where the LDA projection space maximizes the ratio of the between-class variance to the withinclass variance. This special advantage is very important in biometrics field for the classification process since it guarantees maximum class separability [8] LDA lower dimensional space needs three main steps to be calculated. The first one is to find the separability between the different classes (the distance between the means of the different classes), which is represented by calculating the between-class matrix or variance [6] [7]:…”

“…Also, feature extraction techniques can be categorized based on linearity as:  Linear algorithms: these algorithms seek a linear transformation that sets apart different classes. However, if the classes are not linearly separable, linear algorithms fail to find a lower dimensional space where there will be a large overlap between the different classes [6][7] [8]. The rest of this paper is organized as follows: Section 2 states the related works.…”