نوع مقاله : مقاله علمی پژوهشی
نویسندگان
1 استاد، گروه مدیریت صنعتی، دانشکده مدیریت صنعتی و فناوری، دانشکدگان مدیریت، دانشگاه تهران، تهران، ایران.
2 دانشجوی دکتری، گروه مدیریت صنعتی، دانشکده مدیریت صنعتی و فناوری، دانشکدگان مدیریت، دانشگاه تهران، تهران، ایران.
چکیده
کلیدواژه‌ها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Objective
This study seeks to explore and compare ensemble learning models for more accurate predictions of students’ national ranks in Iran’s nationwide university entrance examination, commonly known as the Konkur. The primary aim is to identify optimal models capable of predicting students’ ranks with the highest precision by analyzing data from preparatory and simulated exams conducted before the Konkur. These identified models can then empower academic counselors to offer data-driven recommendations, assisting students in making informed decisions about their educational paths and academic planning.
Methods
Initially, Octoparse software was utilized to collect data related to the preparatory tests conducted by the Kanoon Farhangi Amoozesh (Cultural Center of Education, also known as Ghalamchi) to facilitate a precise analysis of students' performance. The dataset contained key information such as the average test scores, students' national and regional ranks in the exams, as well as the academic disciplines and universities where the students were admitted. Additionally, four advanced ensemble learning models—XGBoost, LightGBM, CatBoost, and Random Forest—were selected for comparison and evaluation based on their performance in predicting students' national ranks. To assess the accuracy and efficiency of these models, various metrics were used, including Mean Squared Error (MSE), Root Mean Squared Error (RMSE), coefficient of determination (R²), and training and prediction times. Afterward, The collected data were split into training and testing sets to ensure optimal model training and evaluation. To achieve the best possible performance, the grid search method was applied, enabling precise tuning of the model parameters to obtain optimal results.
Results
The study findings underscored the exceptional performance of the XGBoost and LightGBM models in predicting students' national ranks. XGBoost emerged as the most accurate model, delivering predictions that are closely aligned with the actual values and exhibiting the lowest error rates, positioning it as the top performer in this research. Similarly, LightGBM, with results closely mirroring those of XGBoost, was also recognized as a strong contender for predicting national ranks. Its blend of speed and precision made it a highly effective tool in this context. The Random Forest model also demonstrated superior accuracy compared to CatBoost, albeit requiring more time for training and prediction. In contrast, CatBoost was identified as the weakest option, displaying lower accuracy and higher error rates than the other models, despite its faster prediction times. This suggests that CatBoost may require further refinement to compete with the more successful models.
Conclusion
The research findings highlight the effectiveness of XGBoost and LightGBM models in ensemble learning, showcasing their exceptional performance in predicting students' ranks in nationwide examinations. With their high accuracy and optimal performance, these models can serve as valuable tools within educational systems, contributing to the improvement of learning processes. They have the potential to identify students who may encounter challenges in their educational journeys and aid in the development of more effective learning programs. Furthermore, the outcomes of this study can assist school leaders and educational policymakers in making well-informed decisions to promote educational equity and ensure equal learning opportunities for all students. Looking ahead, leveraging more advanced deep learning models and integrating additional data, such as social and economic factors and school types, could lead to enhanced prediction accuracy. Additionally, the integration of diverse machine learning models to create hybrid systems may enhance precision and minimize prediction errors. Overall, this research marks a significant milestone in the advancement of predictive and decision-making systems within the education sector, laying the groundwork for substantial progress in this field.
کلیدواژه‌ها [English]
)