MNIST Written Digit Recognition

The goal of this mini project is to use different machine learning algorithms to recognize handwritten digits. The dataset used is the MNIST dataset which is a collection of 28x28 pixel images of handwritten digits. The dataset contains 60,000 training images and 10,000 testing images. The goal is to correctly identify digits from 0 to 9.

The following sections will describe the preprocessing performed followed by the different methods used thus far for this task.

Preprocessing

The dataset is normalised to have values from 0 to 1. The images are also flattened to be 1D arrays of 784 elements.

Methods

Classic Models

These models share a lot of the same logic thus most of their logic is generalised in the ClassicModel class.

1. (Non-)Linear Support Vector Classifiers (SVC)

2. Random Forest

3. K-Nearest Neighbours

Neural Network Models

4. Convolutional Neural Network (CNN)

Evaluations

The evaluation of each model is done in a jupyter notebook found in each model's respective directory. They can be found here:

1. Linear SVC
2. Non-Linear SVC
3. Random Forest
4. KNN
5. CNN