Optimizing Medical School Enrollment

The increasing physician shortage, coupled with an aging population, presents significant challenges for healthcare systems. With higher education facing a projected enrollment cliff, and a decline in youth math and reading scores, identifying the most qualified medical school applicants is imperative. With thousands of applications received annually for only 300 spots at Western University of Health Sciences (WesternU), it is crucial to streamline the selection process while minimizing applicant attrition and melt. In our research project, we develop a predictive model to identify candidates for interviews based on success data from students at WesternU. We utilized a dataset provided by WesternU of 30,000 applicant records from 2018 to 2024, encompassing various demographic, academic, and application metadata. Data preprocessing techniques included removing NaN values, checking for collinearity, applying min-max scaling, and using ADASYN oversampling. Traditional machine learning models yielded accuracies between 35% and 45%. In our final model, our three target variables, 4th year cumulative GPA, COMLEX Level 2, and COMLEX Level 3 test scores, were categorized. For the final model, we decided to employ Linear Programming to calculate feature importances, maximizing mean differences to create a point value ranking system and to employ an ensemble approach. The results show a GPA_CAT accuracy of 36%, COM2_CAT accuracy of 42%, and COM3_CAT accuracy of 55%, leading to a mean accuracy of 44%.