A Comparative Study on Handwritten Digit Recognizer using Machine Learning Technique

Abstract

Ability for accurate digit recognizer modelling and prediction is critical for pattern recognition and security. A variety of classification machine learning algorithms are known to be effective for digit recognition. The purpose of this experiment is rapid assessment of multiple types of classification models on digit recognition problem. The work offers an environment for comparing four types of classification models in a unified experiment: Multiclass decision forest, Multiclass decision jungle, Multiclass Neural Network and Multiclass Logistic Regression. The work presents assessment results using 6 performance metrics: Overall accuracy, Average accuracy, Micro-averaged precision, Macro-averaged precision, Micro-averaged recall and Macro-averaged recall. The experimental results showed that the highest accuracy was obtained by Multiclass Neural Network with the value of 97.14%. The operational tool has been published to the web and openly available at Azure Machine Learning Studio for experiments and extensions.