Deep Learning: Digit Image Recognition
Identifying Hand-drawn Digit Images
This is a Kaggle competition project that involves the building of a prediction model to correctly identify digits from a dataset of ten thousands of handwritten images.
The data files contain gray-scale images of hand-drawn digits ranging from 0 to 9. There are a total of 784 pixels with each image having 28 pixels in height and 28 pixels in width.
The image below shows some examples of the hand-drawn digit images from the data:
The model was built with Convolutional Neural Network (CNN) model of Keras API (Tensorflow backend). My best model gave me an accuracy of 99.7% and placed me in rank 145th out of 2629 total participants (top 6%).
You can access the full project here: