The supervised classification of data in a high-dimensional space is a delicate problem which consists of creating a system
that can predict the correct class of domain objects. It is based on building a procedure that will be applied to a sequence of instances, where each new instance must be assigned to one of a set of pre-defined classes on the basis of observed attributes or features.In this work, we were focused on the recognition ofcharacters, this problem of pattern recognition field , known as OCR (Optical character recognition), remains one of the most popular problems due to its various applications such as forms processing, indexing archives, address classification system, processing of bank check, analysis of written gesture,interaction with the electronic pen, etc. The aim is to transform a text image into an understandable representation for machine and easily reproducible. Many works have been conducted for different languages, an overview of the latest works can be found in [1].Recently, researchers have begun to give attention to the Amazigh language OCR. In this context, various methods have been used for handwritten characters based on: Hidden
Markov Models (HMM) [2], neural approaches [3],geometrical and statistical methods [4][5], moments features
[6] and some hybrid methods [7][8][9]. As mentioned previously, this work focused on a comparison of different supervised classifiers. In this context,we adopted a classification system composed of several steps as shown in figure 1.