A Comparison of Three Classification Algorithms for Handwritten Digit Recognition

Handwritten digits recognition is considered as a core to a diversity of emerging application. It is used widely by computer vision and machine learning researchers for performing practical applications such as computerized bank check numbers reading. However, executing a computerized system to carry out certain types of duties is not easy and it is a challenging matter. Recognizing the numeral handwriting of a person from another is a hard task because each individual has a unique handwriting way. The selection of the classifiers and the number of features play a vast role in achieving best possible accuracy of classification. This paper presents a comparison of three classification algorithms namely Naive Bayes (NB), Multilayer Perceptron (MLP) and K_Star algorithm based on correlation features selection (CFS) using NIST handwritten dataset. The objective of this comparison is to find out the best classifier among the three ones that can give an acceptable accuracy rate using a minimum number of selected features. The accuracy measurement parameters are used to assess the performance of each classifier individually, which are precision, recall and F-measure. The results show that K_Star algorithm gives better recognition rate than NB and MLPas it reached the accuracy of 82.36%.