| Summary: | Data Envelopment Analysis (DEA) is effective in evaluating efficiency across various domains, including education. However, traditional DEA methods face challenges such as computational complexity and limited predictive capabilities. This study addresses these limitations by integrating Automated Machine Learning (AutoML) using Genetic Programming (GP) with DEA to enhance the prediction and efficiency analysis of academic performance among final year diploma students. The objective is to develop and evaluate an integrated framework that leverages AutoML and DEA to improve the accuracy and generalization of efficiency predictions. The study investigates the impact of varying population sizes, mutation rates, and crossover rates within the AutoML framework on model performance. Methodology involves collecting data from 1,099 final year diploma students through surveys measuring their CGPA, satisfaction, and various competencies. Four CCR and BCC models were constructed using different input-output combinations, with Model BCC1 identified as the best model. Cross-validation was employed to mitigate overfitting. Result revealed increasing population sizes generally improved performance, with error values decreasing for both datasets. The best-performing pipelines included RandomForestRegressor and XGBRegressor model by highlighting their robustness and adaptability. In conclusion, integrating GP-based AutoML with DEA provides a promising approach to enhancing efficiency analysis of academic performance. Future research should focus on addressing overfitting through other strategies such as regularization and extending this methodology to other domains for further validation and improvement. © 2024 IEEE.
|
|---|