Yapay Zeka Hızlı Bakış & Çıkarımlar
AI InsightStatistical analyses (paired t-test and Wilcoxon test) confirmed that ensemble models significantly outperformed traditional ones in terms of both accuracy and F1-score (p < 0.05).
📌 3 Maddede Bu Makale
Abstract
On-time graduation is one of the primary goals of undergraduate programs, aiming to equip students with the necessary knowledge and skills in a specific field to ensure their rapid integration into the workforce. However, delays in graduation can postpone career entry, increase financial burdens, and create psychological stress. Therefore, accurately predicting students' graduation outcomes at early stages is critically important for higher education institutions to develop effective educational strategies and provide timely interventions for at-risk individuals. This study aimed to determine the most effective approach for predicting on-time graduation by comparing ensemble learning methods with traditional machine learning models using demographic data, high school performance, and progressively accumulated university academic data over six academic periods. The models used include ensemble methods such as Random Forest, CatBoost, and XGBoost, as well as traditional models like Logistic Regression, Decision Trees, and Support Vector Machines. Across seven prediction checkpoints (T0–T6), CatBoost consistently outperformed all other models, achieving 83.0% accuracy, 84.8% F1-score, 84.4% precision, and 85.1% recall in the final prediction stage (T6). Among traditional models, Logistic Regression performed the best, with 83.5% accuracy and 84.7% F1-score at T6. Statistical analyses (paired t-test and Wilcoxon test) confirmed that ensemble models significantly outperformed traditional ones in terms of both accuracy and F1-score (p < 0.05).