Akhila

Akhila

MSCS, UIUC

Use of Machine Learning and Deep Learning Models with Longitudinal EHR to Better Predict 10-year Stroke Risk

  • Github Repo
  • Designed a classification problem in collaboration with a 2-person student team and domain experts at the Department of Biomedical Informatics (DBMI), VUMC
  • Utilized and evaluated innovative machine learning algorithms from decision trees to neural networks in Python to classify outcomes and compare algorithm performance with traditional stroke risk prediction models; most balanced model produced an accuracy of 72.4% and a false negative rate of 59.0%
  • Analyzed technical problems surrounding algorithms to troubleshoot major technical issues and optimize machine learning models while testing the best hyperparameters