Predicting University Rankings Using Random Forest Regression on Institutional Metrics: A Data Mining Approach for Enhancing Higher Education Decision-Making

Abstract

This study investigates the prediction of university rankings using Random Forest regression, leveraging institutional metrics as input features. The primary objective is to enhance the decision-making process in higher education by providing a data-driven model capable of forecasting rankings with greater transparency and accuracy. The research utilizes a comprehensive dataset containing institutional metrics such as research quality, teaching effectiveness, international outlook, and industry impact. Random Forest regression is chosen for its robustness, handling both linear and non-linear relationships between features and the target ranking variable. Feature selection techniques, including correlation analysis and dimensionality reduction, are applied to identify key metrics that influence rankings. Through rigorous model training and hyperparameter tuning, an optimal Random Forest model is developed indicating strong predictive accuracy. Evaluation metrics such as Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R² are used to assess model performance. The feature importance analysis reveals that research quality and research environment have the highest impact on university rankings, followed by teaching and international outlook. These findings align with common assumptions in higher education rankings, while also revealing the potential of less-studied metrics, such as industry impact and international student population, to influence rankings. This study contributes to the field of open education by presenting a transparent and accessible method for predicting university rankings. It empowers students, administrators, and policymakers with a data-driven approach to assess institutional performance. The research also highlights the limitations of current ranking systems and suggests avenues for future studies, including the use of multi-year datasets and alternative machine learning models. , , , ,