Handwritten Numerals' Recognition in Kurdish Language Using Double Feature Selection

In the contemporary rapidly developing world, machine learning and pattern recognition has achieved a special level of attention by the academicians and international researchers. Such methods have economized time and increased the speed of processing tons of data in a single click. Pattern recognition is being used in a broad context, especially in industry, finance and banking, forensic issues, etc. Even though pattern selection programs have economized our time in an eye-catching way, yet choosing the best and most accurate algorithm is still a challenge for many researchers. In this paper, a Double Feature Selection (DFS) method as a new trend, using Correlation and ReliefF, is being elaborated in order to appraise the ratio of digit recognition accuracy, instead of using them individually. For this purpose, a comparison is implied to illustrate the accuracy of the four algorithms; Decision Tree (J48), KNN, Multi-Layer Perceptron (MLP), and Naïve Bayes (NB) as far as they are concerned with the digits and numerals used in Kurdish Language. The result of the study illustrates that DT has the best accuracy with the ratio of 95.6%.