Comparing an Integrated Data Envelopment Analysis and Machine Learning Models for Accurate Academic Efficiency Prediction

Abstract

The integration of data envelopment analysis (DEA) with machine learning (ML) offers a novel approach to evaluating academic efficiency beyond traditional measures like CGPA. This study develops an efficiency assessment framework combining DEA and ML to predict student academic achievement efficiency. The objectives are (1) to identify and validate input and output variables for DEA-based academic efficiency measurement and (2) to develop an integrated predictive model using DEA and ML for improved accuracy. A cross-sectional study was conducted on 1,099 final-year diploma students, collecting data on CGPA, satisfaction, and five competency domains (personal, adaptive, digital, social, and 21st-century skills). Efficiency scores were computed using the BCC and CCR DEA models, followed by ML predictions using random forest (RF), gradient boosting regressor (GBR), artificial neural networks (ANN), and AutoML via genetic programming. Performance was evaluated using RMSE, MAE, and R² metrics. The findings indicate that the DEA-GBR model achieved the highest predictive accuracy (RMSE = 0.0101, MAE = 0.0039, R² = 0.9889), outperforming other models. SHAP analysis identified digital competency as the most influential predictor, aligning with UiTM’s digital transformation goals. The integration of DEA with ML significantly improved discriminatory power, reducing the number of efficient decision-making units (DMUs) from 134 to as low as 44. This study enhances academic efficiency assessment by integrating DEA with predictive ML models, providing a data-driven approach for student performance evaluation. Future research should expand datasets and explore additional ML techniques for further refinement.