Performance Evaluation of Machine Learning Algorithms in Handwritten Digits Recognition

the recognition of handwritten characters has always been a very difficult task because of the many variations of handwritten characters with different writing styles. This type of intelligent systems is applied in various fields: check processing, processing of forms, automatic processing of handwritten answers to an examination, etc. This last application is the subject of this work. We compared in this paper the performances of some machine learning algorithms, used for the classification of complex and multiclass problems. In this work, we exploited four machine learning algorithms (K-Nearest Neighbors, Deep Neural Network, Decision Tree and Support Vector Machine) to predict handwritten digits. The training and testing data were extracted from the MNIST digit database containing pre-processed images. The results obtained using different similarity measures such as accuracy, sensitivity and specificity confirm that the classification obtained by deep neural networks is the most accurate compared to the other classifiers studied in this paper.