Early detection of at-risk health sciences students: a machine learning-based predictive study using midterm grades

Area of Science:

• Health Sciences Education
• Machine Learning Applications
• Academic Analytics

Background:

• Early identification of at-risk students is crucial in health sciences education, especially for healthcare workforce development.
• This study explores machine learning (ML) classifiers for this purpose in the United Arab Emirates (UAE).
• Midterm performance was used as an early indicator of academic integration and student persistence.

Purpose of the Study:

• To evaluate the effectiveness of established ML classifiers in identifying at-risk health sciences students in the UAE.
• To assess the utility of first-term midterm performance as a predictive indicator.
• To provide a regional case study for academic intervention strategies.

Main Methods:

• Analysis of academic records from 346 first-year students across seven programs and three common courses.
• Training of five supervised ML models (XGBoost, Random Forest, Logistic Regression, GBM, Naïve Bayes) using ROSE with five-fold cross-validation.
• Performance assessment using F1-scores and AUC, with interpretability via variable-importance plots and SHAP for XGBoost.

Main Results:

• XGBoost and Logistic Regression models demonstrated the highest performance (F1=0.84; AUC>0.91).
• Midterm scores in Biology and Introduction to Health Sciences were the strongest predictors of academic risk.
• SHAP analyses confirmed the significant impact of these midterm scores on risk predictions.

Conclusions:

• Midterm grades serve as feasible and pedagogically meaningful early indicators for identifying at-risk students in UAE health sciences programs.
• The study offers regionally relevant evidence to support timely academic interventions.
• Future research should incorporate behavioral/psychosocial factors and validate findings across diverse cohorts and institutions.