A Time-Adaptive Ensemble Framework for Multi-Year University Ranking Prediction Integrating Outlier-Aware Scoring and Hybrid Feature Selection

Muhammad Aria Rajasa Pohan (Department of Electrical Engineering, Universitas Komputer Indonesia, Indonesia)
Bagus Abdul Muiz (Department of Electrical Engineering, Universitas Komputer Indonesia, Indonesia)

International Journal of Informatics, Information System and Computer Engineering (INJIISCOM)
Vol. 7 No. 1 (2026): INJIISCOM: VOLUME 7, ISSUE 1, JUNE 2026 (ONLINE FIRST)

University ranking prediction requires adaptive models to capture temporal dynamics and handle data anomalies. This study develops a time-adaptive ensemble framework integrating outlier-aware scoring and hybrid feature selection. Using Times Higher Education data (2011–2024), we applied windowed outlier detection with clipping and masking, alongside ANOVA, permutation importance, and SHAP values for dynamic feature selection. The framework ensembles linear moving-average, temporal Random Forest, and LSTM models via optimized weights. Rolling forecasts (2016–2024) yielded a low mean rank deviation of 1.2 positions and a Top-1000 classification accuracy of 0.96, outperforming single baselines. This robust, interpretable framework effectively supports strategic decision-making and resource allocation in higher education