oa An experimental comparison of supervised classifiers on two real datasets for pattern recognition

In this paper, we present a comparative study between Support Vector Machine (SVM) and Adaboost, as being two decision-based classification tools in the field of shape recognition. The aim of this work is to study their theoretical foundations, their learning algorithms and to investigate their performance in classification capacity. To compare their performance, we apply them to two famous training datasets, is widely used by the community, namely the CBCL- MIT face and Wisconsin diagnosis breast cancer (WDBC). The quality of decision of each classifier depends on the choice of its parameters and its implementation.

The field of pattern recognition [1] has witnessed a revolution since the mid-90s with the statistical learning theory and the advent of the Support Vector Machines [4] (SVM) for the resolution of detection problems, classification and regression. In recent years, a set of interdependent disciplines, concerning the information treatment, decision theory and methods of pattern recognition, Boosting methods [7] has emerged. The applications of pattern recognition are extended to include several areas such as shape recognition, the approximation of functions, image processing, speech recognition, and classification. The objective of this paper is to compare their performance in the field of the supervised classification.

The term may refer to two classes of distinct methods: the supervised classification and unsupervised classification. Non supervised methods are intended to constitute examples groups (or groups of instances) based on the observed data, without a priori knowledge. On the other hand, supervised methods use a priori knowledge on the belonging of a sample to a class to build a recognition system based on these classes. In this paper we focus on supervised classification.

The goal of supervised classification is to build, using a set of training data (training set), a classification procedure which allows predicting membership of a new example to a class. Our goal in the near future is to continue the study of SVM and Adaboost in order to test the relationships that exist between them.