Predicting High Research Impact Universities Using Bibliometric Indicators: A Machine Learning Study

University research impact is commonly measured using bibliometric indicators based on citations; however most existing studies focus on using these indicators to analyse the present or past research outcomes, leaving a gap in the predictive use of these indicators. In this study we have addressed this limitation by employing a temporal prediction framework which predicts high research impact status in the$\mathbf{T}+\mathbf{1}$period based on bibliometric indicators from the period T. The CWTS Leiden Ranking Open Edition 2025 data of 2,831 universities was used for this purpose. This prediction task was formulated as a binary classification problem, and the performance of logistic regression was compared with ensemble-based machine learning models, including Random Forest and XGBoost. Results show that ensemble models achieve superior discrimination performance, with XGBoost attaining an ROC-AUC of approximately 0.87 and also achieving a better F1-score. Feature importance analysis shows that publication volume and international collaboration are the strongest predictors of future high research impact status. Overall, this study demonstrates the potential of ensemble-based machine learning methods for prediction of future university research impact using transparent bibliometric indicators.