Vanderbilt DSI and Women’s Basketball team collaborate using data science to improve performance while reducing risk of injury
The Vanderbilt University Data Science Institute continued its project with the Vanderbilt Women’s Basketball strength and conditioning program to use data to improve athletic performance while reducing the risk of sport-associated injury. This semester, the focus was on making better use of training and competition data through improved analytics efforts, data consolidation, the creation of predictive models, and improved data visualization.
The data science team working on the project this semester included Umang Chaudhry, Grace Ko, Minwoo Sohn, Nikkhil Niranjan, and Sovann Chang. The team also continued to work on the project with Tyler Clarke from Vanderbilt Athletics.
The project utilizes data from various sources, including Catapult GPS technology, Vald: Force Decks dual force plate technology, Vald: NordBord hamstring testing systems, and Fusionetics assessment protocols. Last semester, the team worked on understanding the data collected by the team, getting the APIs to work, and beginning work on creating a dynamically changing and de-identified dashboard.
This semester, the team worked on challenges with the APIs themselves as they incorporated them into the dashboard, pulling the necessary data, and detecting anomalies. Anomaly detection was a key area of focus this semester, with the goal of finding any indications of injury in each athlete’s data and alerting staff if similar signals are found to prevent injury.
The team utilized Variational AutoEncoders (VAE) for anomaly detection, which detects anomalies of each metric of each player. Over the summer, they will decide which anomalies will be considered by each metric and finish up the code. A report of possible signals of injury will also be created.
Additionally, the team worked on pulling more recent data and modifying the daily pull function to make it more seamless. The dashboard took a while to load because it was pulling in data that was months old, but now it pulls in data that is more recent and useful for training purposes. However, the team faced a dilemma with dashboard hosting, as their data is pulled from various sources and APIs, and most of their packages are sourced from Github/CRAN, which are compatible with Shinyapps.io. They experienced some difficulties with pulling in Catapult data into the dashboard because it was on Bitbucket and not Github, which made their dashboard incomplete and not fully comprehensive. Therefore, the dashboard is currently hosted locally.
The next steps for the project include optimizing, enhancing, and deploying the dashboard, implementing anomaly detection across athletes and metrics, and incorporating it into the dashboard. The team also plans to create predictive models for ForceDecks, Nordbord, and Fusionetics metrics using Catapult metrics.
Read the original full story here.
