Abstract
In this paper, a machine learning approach for classifying handwritten Arabic word is proposed, which includes three stages including preprocessing, feature extraction and classification. Firstly, words are segmented front inputted scripts and also normalized in size. Secondly, three different feature extraction methods are applied to each segmented word namely the discrete cosine transform (DCT), moment invariants, and absolute mean value of overlapping blocks. Finally, theses features are utilized to train a neural network jot classification. This approach has been tested using the IFN/ENIT database which consists of 32492 Arabic words. The proposed approach gives a good accuracy when compared with other methods.