Machine Learning - Driven Student Ranking for Efficient University Accommodation Allocation

The allocation of limited university accommodation in Bangladesh is currently a manual, labor-intensive process prone to human error and subjective bias. While administrative decisions rely on both quantitative metrics (e.g., CGPA, income) and qualitative appeals (e.g., student “Family Situation” narratives), existing automated systems rarely process both simultaneously. To address this, we present a novel, manually annotated dataset comprising 3,454 real-world applicant profiles. We introduce a multimodal machine learning framework that utilizes this dataset's seven distinct attributes-fusing quantitative metrics (e.g., CGPA, Income, Disciplinary Record) with unstructured qualitative appeals-to automate student ranking, quantitatively evaluated for algorithmic fairness across socioeconomic subgroups. We implement a data preprocessing pipeline that standardizes multi-lingual application narratives into a unified English format, enabling the extraction of text features via TF-IDF. While TF-IDF was selected for its computational efficiency and high interpretability in an administrative context, it serves as a robust baseline for capturing key indicators of student hardship. These textual features are fused with socio-economic and academic indicators to train a priority-ranking model. A comprehensive evaluation of 28 algorithms reveals that Random Forest Regression achieves the highest fidelity, replicating human expert decisions with 89% accuracy. By converting subjective student narratives into objective, quantifiable data, this system offers a scalable, transparent, and data-driven approach to resource allocation in higher education.