Recognition of Handwritten Digits with the Help of Deep Learning

Handwritten numerical recognition is becoming the most interesting topic in research area today due to great growth in artificial intelligence and its different learnings and computer visual perception algorithms. This project shows the comparison of digit recognition among machine learning algorithms like support vector machine (SVM), K-nearest neighbor (KNN), random forest classifier (RFC) and with deep learning algorithm like multilayer convention neutral network (CNN) using Keras (Keras is a high-level neural networks library written in Python which is simple enough to be used. It works as a protector to low-level and high-level libraries like TensorFlow or Theano) with Theano and Tensorflow (An open source software library which provides high performance numerical computation. Its architecture is flexible in such a way that it is easily deployed across various platforms like (CPUs, GPUs, TPUs), form desktop to clusters of servers to smart handsets devices). Further looking and comparing for the accuracy produced by above-mentioned algorithms, the results appear to be like this: The accuracy of digit recognition is 98.69% in convolutional neural network, 97.90% in support vector learning (SVM), 96.67% using K-nearest neighbor (KNN) and 96.89% using random forest classifier (RFC) which clearly shows that convolutional neural network produces more accurate prediction with better results comparatively.