Advanced Methods for Data Science

Coursework completed as part of the Methods for Data Science course at Imperial College London.

Exploration of two different high-dimensional data sets - the MNIST-Fashion dataset, and a dataset of text documents with citations. The approaches considered are:

• Supervised classification (Multi-Layer Perceptron, Convolutional Neural Networks, K-Nearest Neighbours)
• Graph analytics (Community Detection)
• Unsupervised learning (K-Means Clustering)
• Dimensionality reduction (Principal Component Analysis, Non-Negative Matrix Factorisation, Latent-Dirichlet Allocation)

Sections

1. Unsupervised Learning

• K-Means Clustering
• Graph Analytics - Community Detection and Centrality Measurements
• Comparison between Communities and Clusters

2. Supervised Classification

• K-Nearest Neighbours
• Multi-Layer Perceptron
• Convolutional Neural Networks
• Evaluation and Comparison

3. Poster

• A poster visualising and explaining results.

4. Dimensionality Reduction

• Principal Component Analysis
• Non-Negative Matrix Factorisation
• Latent-Dirichlet Allocation