Data input issues, which seem to be a barrier for the data computing sector, could be solved with an Optical-Character-Recognition (OCR) mechanism. As a result, OCR mechanisms have already been designed virtually for all languages in the world, including Arabic. Over the last 30 years, much fundamental research has gone into the creation of an effective Arabic-OCR (AOCR) mechanism. Because of the increasing quantity of materials accessible on the website, in email accounts, and online databases, documentation categorization has become a necessary job. It's usually accomplished following proper selection of features, which entails choosing suitable characteristics to improve the accuracy of classification. The large percentage of feature-based textual classifying techniques depend on creating a term-frequency and inverse-documentfrequency based on features representation, which is inefficient across many cases. Furthermore, many content categorization research is concentrated on the English-language. Despite the difficulty of the Arabic-language, this research paper concentrates on AOCR, which has received less attention. The" Extended Particle Swarm Optimization"(EPSO) methodology is introduced for selecting optimal features from extracted features to comply with Arabic-Character-Classification, and the" Enhanced K-Nearest Neighbor"(EKNN) has been employed as a classification model for identifying and classifying or simply recognizing the particular Arabic character in this research article. Various assessment metrics, comprising Accuracy, Precision, and Recall, are utilized to evaluate this method. Experimentation on an actual dataset is also carried out, and also a comparative with existing Deep-Belief-Networks (DBN) techniques had performed.