Predicting Students at Risk of Dropping Out in the College of Medicine at Mohammed Bin Rashid University of Medicine and Health Sciences. A Machine Learning Approach to Understand Attrition Factors Over the Past Seven Years

This master’s thesis investigates student attrition within the College of Medicine at Mohammed Bin Rashid University of Medicine and Health Sciences (MBRU) by applying the CRISP-DM methodology to analyse a dataset of 19 variables and around 2,000 records from Student Admission and Academic Performance sources. Although the dataset was relatively small, several machine learning models were developed to predict students at risk of dropping out, including Logistic Regression, Decision Trees, Support Vector Machines (SVM), and Artificial Neural Networks (ANN). Of these models, the ANN model, particularly when combined with Principal Component Analysis (PCA), achieved the highest performance with an accuracy of 97%, a recall of 94%, a precision of 83%, and an F1 score of 88%. The findings revealed that academic performance measures, including Year 1 Grade Point Average (Y1_GPA), Year 2 Cumulative GPA (Y2_cGPA), Year 3 Cumulative GPA (Y3_cGPA), and whether a student repeated a year (Repeat Year), were significant predictors of student attrition. These academic factors accounted for approximately 63% of the model's predictive power. Additionally, admission-related variables, such as the Entrance Exam and Multiple Mini Interview (MMI) scores, contributed around 15% to the model's predictive accuracy. These findings emphasize the importance of developing a multidimensional dataset for a more holistic analysis of student attrition. The research adds to the limited body of studies on the use of machine learning in the UAE and GCC, and aids in the ongoing adoption of AI-based solutions at MBRU, particularly for enhancing student retention strategies, demonstrating the broader potential of these technologies in education.