Reuben College hosted a successful inaugural datathon on Saturday 11th June that included graduate students from Reuben, Linacre, Keble, and Green Templeton Colleges. By the end of the day all students independently developed reproducible machine learning models to classify activities of daily living from wrist-worn activity trackers.
The session began at 9:30am, with bright eyed students filtering into the common room. Attendees came from a variety of academic backgrounds. This included some that worked in theoretical machine learning, who were seeking to discover the application of these techniques. Most, however, came from a background in health related fields, without much experience in machine learning. Following introductions, the session started with presentations delivered by Aiden Doherty, introducing the students to the basics of physical activity monitoring, and then Shing Chang, covering the processing of wearable accelerometry data using machine learning.
After these presentations, attendees had the opportunity to develop hands-on machine learning skills, supported by Aidan Acquah and Shing Chang. All were able to get machine learning code set up on the laptops, and ran through various stages of data processing, attempting to answer various questions along the way. The main focus of the supervisors was in supporting the attendees in answering the questions, particularly noting the difficulties in machine learning pre-processing and feature engineering. After time was given for attendees to run through various tasks, Aidan Acquah presented solutions to the exercises.
After completing a run through of all the exercises, attendees were split into two groups, to attempt to improve the feature extraction for activity recognition. Attendees proposed alternate features, to improve the performance of the machine learning model provided. After learning the importance of physical activity for cardiovascular health, the day included a group walk through University Parks to ensure practical maintenance of cardiovascular health as well as of coding skills! At the end of the day, attendees shared progress of their machine learning classifiers with the wider group over dinner.
Some quotes from students who took part in the datathon:
I would say that, from the perspective of someone working in AI theory, the Datathon was a refreshing opportunity to get insight on the practical challenges of applying machine learning to an exciting and impactful wearables dataset. The organisers were excellent and helped to foster discussion between all participants regardless of experience. - Benjie
I thought the datathon was an excellent experience to interact with a real-world healthcare dataset. The team was very prepared with activities for a range of ability levels, and was able to provide one-on-one guidance if any of us got stuck along the way. - Klara
