Predicting Concussion Symptom Resolve Time in High School Athletes  
Author Sara Landset


Co-Author(s) Taghi M. Khoshgoftaar


Abstract Sport-related concussions can be very serious injuries, particularly among young people whose brains are still developing, and individualized treatment plans are necessary in order to ensure the best outcomes. Proper development of such a plan would benefit from an effective method of predicting the amount of time it will take until all symptoms are resolved. We have previously proposed a machine learning approach to this problem for injuries suffered by high school students playing contact sports, examining 10 learning algorithms and three symptom resolve time frames. In this study, we demonstrate that those predictive models maintain their good performance when we expand the dataset to include all sports, despite the introduction of more variability to the data. However, this appears to only be the case provided there are enough training instances to learn from. Reducing the dataset to only noncontact sports degrades the performance of most models, and it is unclear whether this is caused by the added variability, the reduced number of instances, or both. Additionally, we find that models predicting shorter symptom resolve times perform better than longer ones, and among the 10 learners, Na¨ıve Bayes is the most reliable on all versions of our dataset.


Keywords Concussion, Sports Medicine, Traumatic Brain Injury, Machine Learning, Return-to-Play, Classification.
    Article #:  DSIS19-40
Proceedings of ISSAT International Conference on Data Science & Intelligent Systems
August 1-3, 2019 - Las Vegas, NV, U.S.A.