A Comparison of Machine Learning Techniques for Yemeni Universities Admission Examinations Predictions

Abstract

Admissions examinations are crucial in shaping students' academic and career paths. Machine learning (ML) offers a promising avenue for transforming the assessment process in universities, schools, and colleges. This paper presents a new ML-based method to evaluate high school graduates seeking admission to eleven colleges across nine Yemeni universities, applicable to both scholarship and self-financing students. The proposed approach leverages several ML techniques, including Support Vector Machine (SVM), Logistic Regression (LR), Random Forest (RF), K-Nearest Neighbors (K-NN), Gradient Boosting (GB), Extra Trees (ET), Decision Tree (DT), AdaBoost, and XGBoost. In this study, ML algorithms were applied to admissions examinations at nine Yemeni universities, including Sana'a, Ibb, Dhamar, Amran, Jiblah, Mihwit, Al-Dhalea, Saada, and Hajjah. The dataset comprised 15,898 student applications for eleven colleges, encompassing both scholarship and self-financing programs. Experiment one evaluated the performance of various ML techniques. Gradient Boosting (GB) and Random Forest (RF) demonstrated superior results, achieving the highest accuracy, precision, recall, and F-score. Specifically, GB and RF attained 91.00% and 90.00% accuracy, 87.60% and 88.80% precision, 89.90% and 87.50% recall, and 89.30% and 87.60% F-score, respectively. Experiment two further refined the analysis, comparing ET and RF to other ML methods. ET and RF consistently outperformed LR, KNN, SVM, DT, GB, AdaBoost, and XGBoost. In this experiment, ET and RF achieved exceptional accuracy, precision, recall, and F-score rates of 99.00% and 99.00%, 97.50% and 97.20%, 98.00% and 97.80%, and 97.80% and 97.50%, respectively.