Simple MNIST Digit Classifier Neural Network

In this research project, I implemented a Simple MNIST Digit Classifier Neural Network that recognises hand-written numerical digits from the MNIST Digit Recogniser Dataset made from Scratch* in Python with 7960 trainable parameters using Forward Propagation, Back propagation, ReLU & Softmax Activation Functions…

This research paper presents an implementation of a basic feed-forward neural network for classifying/recognising handwritten digits from the MNIST dataset, built from scratch in Python using only core libraries such as NumPy, Matplotlib, Pandas, and tqdm. This feed-forward neural network model, also known as a multilayer perceptron (MLP), achieves a classification probability accuracy of approximately 89% for both the training and testing datasets. This demonstrates the viability of simple neural networks that recognise handwritten numerical digits without the need for external heavy libraries such as PyTorch or TensorFlow, instead utilising only core Python libraries and Juypter Notebook.