Explainable Health Data Analytics with Tsetlin Machines
Project Overview
This project is focused on using hospital records to train Tsetlin Machines and evaluate their performance against established machine-learning approaches under consistent pre-processing conditions. The project aims to investigate whether Tsetlin Machines can maintain robust predictive performance in the presence of missing clinical data, while providing interpretable clause-level explanations through transparent Boolean logic.
The wider aim is to assess the suitability of Tsetlin Machines as an interpretable and robust approach for predictive modelling using real-world clinical data. This will help evaluate potential implementation in clinical settings in which reliable predictions and an understanding of decision influencing features behind the predictions are important.
What were the key results of your research project?
• I successfully set up pre-processing measure to handle datasets of ranging sizes and missingness to ensure Tsetlin Machine Compatibility.
• I trained 3 different Tsetlin Machine models across datasets with moderate to good performance outcomes in comparison to some standard comparison machine learning models.
• Each Tsetlin Machine provides the clauses that have lead to the machines final predictive decision in a readable format allowing understanding of the decision.
How do you feel you have benefitted from completing this internship and has it made you consider future career paths?
Before this internship, I had little knowledge of how data analytics works in practice or how to apply coding skills developed at university to industry settings and real-world problems. Throughout the project, I have developed my resilience and ability to handle unexpected challenges that extend beyond coding errors, including adapting to unfamiliar software, working with publicly available code and independently identifying solutions to problems.
Additionally, having no prior experience of research or data-driven projects, this internship has been particularly eye-opening and has encouraged me to consider research and data analysis roles in the future. I have enjoyed working with a level of supervised independence, where I have been responsible for investigating problems and reaching results rather than having easy access to predefined solutions or answers.
Being able to present my work at the N8 CIR York Showcase is an opportunity I am grateful to have had, and it encouraged me to develop my communication skills alongside my technical abilities. Presenting a technical project within a limited time and to an audience who may not have experience in the topic was challenging, but taught me the importance of communicating complex ideas in a clear and accessible way.