The Performance Evaluation of Deep Learning Classifier to Recognize Devanagari Handwritten Characters and Numerical

Abstract

A text classification is a well formed process using various measurable properties and computerized logical procedure to fetch a pattern from different classes.Since classification is important for the pattern recognition process, there are some issues with well-formed classification in this process, which is one of the important issues for proper development and improvement of productive data examinations. On behalf of the versatility of learning and the ability to deal with complex calculations, classifiers are consistently best suited for design patter recognition issues. The aim of this paper is to present a result based comparative study of different classifiers and the optimal recognition of results computation through the Devanagari Handwritten characters and numerical values. Different classifiers were used and evaluated in this investigation including k-Nearest Neighbor (k-NN), Support-Vector machine (SVM), Naïve Bayes, Decision Tree, Random Forest, and Convolution Neural Network (CNN). To accomplish the experiment purpose, this paper used an unbiased dataset with including 123 samples that consists of 123 characters and 123 numerical values. Python 3.0 with sciket learn machine learning open-source environment library have been used to evaluate the performance of the classifiers. The performances of the classifiers accessed by considering the different matrices including dataset volume with best split ratio among training, validation, and testing process, accuracy rate, Ture/False acceptance rate, True/False rejection rate and the area covered under the receiver operating characteristic curve. Similarly the paper shows the correlation of the accuracy of the experiments obtained by applying to chosen the classifier. On behalf of the exploratory results, the infallible classifiers considered in this test have free rewards and must be executed in a hybrid manner to meet the thigh precision rates.In the views of test work, their result compressions and the examination to be performed, it is argued that the Random Forest classifier is performing in a way that the current use of the classifier to recognize the Devanagari Handwritten character and the numerical values with the accuracy rate 87.9% for the considered 123 samples.