Module Aims

This module aims to link the fundamental concepts presented in “Introduction to Machine Learning” to practical examples frequently encountered in Health Data Science and, in parallel, introduce some advanced elements of previously discussed canonical methods. 

Module Learning Outcomes

By the end of the course, students should be able to: 

1. Describe and evaluate several canonical machine learning methods and feature selection processes (including assumptions, algorithms and examples) 

2. Identify and apply appropriate machine-learning methods (and critically compare the stability of results obtained using standard approaches) to solve a range of inferential and prediction problems 

3. Recognise contexts in which versions of algorithms for inference and prediction derived from flexible high dimensional models can offer advantages over classical implementations or statistical methods 

4. Have a wider understanding of the information which can be derived from preliminary data mining, reinforcement learning or heuristics and apply flexible modelling on real world examples 

5. Identify useful objective function penalisations corresponding to a variety of structural assumptions 

6. Interpret the output of machine learning algorithms in the context of the underlying modelling assumptions. 

 Pre-requisites

• Statistics for HDS
• Introduction to Machine Learning
• Advanced Biostatistics for HDS

Teaching Strategy

Lectures and computer practicals. Some preliminary reading may be required. 

Assessment

Practical analysis and report with two-components. For part 1, worth 50% of the module grade, students will employ various data mining approaches to determine the features of the provided datasets (e.g. noisy data but fully labelled, data with missing values, partially labelled data, sequencing data, imaging data), and write down their comments.   
For the second part of the assignment, students will answer a predefined set of questions comprising the assessment of 2-3 ML approaches (25%) and an open-ended analysis on the results obtained from applying one of the ML approaches on the given dataset (25%). 

 Module Length and Dates

4 days