MNIST Handwritten Digit Recognition Using a Deep Learning-Based Modified Dual Input Convolutional Neural Network (DICNN) Model

The key aspect of user authentication is handwriting. In the era of information technology, the recognition of handwritten numbers has recently become an important aspect. The ability of a machine to detect handwritten digits that are collected through numerous sources like as papers, touch screens, and pictures, and finally categorize these digits into number groups is called human handwritten digit recognition. Various classification approaches like machine learning (ML) and deep learning (DL) algorithms are used for recognizing the handwritten digit. The performance metrics Accuracy-(ACC) for a number of correct predictions, F1-score-( F1) for class-wise performance, Recall-(REC) for the number of positive predictions in the entire dataset, and Precision-(PREC) for the number of positive prediction of the correct model are used to evaluate the various fundamental machine learning algorithms, such as K-Nearest Neighbors (KNN), SVM, LR, NN, RF, NB, and DT algorithms having the accuracy level between 74 and 97%. In this paper, we have proposed and designed a deep learning-based modified "Dual-Input Convolutional Neural Network (DICNN)" model to improve accuracy. We have used a public MNIST dataset of 70,000 samples of handwritten digits. Finally, we have compared the performance with the fundamental machine learning algorithms.