HANDWRITTEN DIGIT RECOGNITION USING ARTIFICIAL NEURAL NETWORK

Abstract

Neural network-based handwritten digit recognition is an important task in computer vision and machine learning. The goal of this research is to create a productive system that can recognise handwritten numbers from pictures with accuracy. The project makes use of the MNIST data set, which is a collection of handwritten numbers that are 28 by 28 grayscale representations of numbers between 0 and 9. To improve the calibre and variety of the training data, the procedure starts with preprocessing the dataset. This includes operations like scaling, normalisation, and data augmentation. Following that, other neural network architectures are investigated, with a particular emphasis on convolutional neural networks (CNNs) because of their superior ability to represent spatial hierarchies in visual data. Next, the chosen model is trained with optimisation algorithms such Adam, or stochastic gradient descent (SGD), is used with hyperparameter adjustment to maximise efficiency. The model that was trained is evaluated on an alternative test set in order to ascertain its accuracy, precision, recall, and F1 score. Lastly, a real-world application utilising the model is implemented, maybe with a user interface that allows users to submit handwritten numbers for prediction. The model's performance and dependability in production are guaranteed by ongoing maintenance and monitoring. With possible applications in areas including optical character recognition, automated document processing, and digit-based authentication systems, this study advances the field of digit recognition technology.