Handwritten digit recognition (HDR) is the rudimentary principal behind today’s generalised optical character recognition. HDR has been applied throughout numerous domains from the sorting of mail, interpreting numerically rated survey responses and reading student candidate numbers on examination scripts.
Whilst HDR has already been extensively analysed, handwriting quirks transition over time. This means a constant ongoing analysis is required to ensure accuracy rates in HDR keep pace with a technologically demanding world.
This paper aims to critically evaluate two supervised learning techniques, Multi-Layer Perceptrons and Convolutional Neural Nets, to classify an image into one of the ten numerical digits. We will conduct a systematic hyper-parameter grid search on both forms of neural networks in order to ascertain our leading algorithm.