MACHINE LEARNING-ENHANCED QUOTA ALLOCATION: A FAIRNESS-AWARE FRAMEWORK FOR CHINA'S HIGHER EDUCATION ADMISSIONS

Abstract

This study implements and verifies a fairness-aware algorithmic framework optimised for machine-learning-powered quota-based equity resolution that specifically addresses systemic inequities in China's higher education admission policies. This research characterises quota allocation as a constrained multi-objective optimisation problem where both efficiency and equity with respect to distribution across allocation groups are measured through demographic parity and equal opportunity constraints. The proposed framework incorporates ensemble learning approaches of Random Forest, Gradient Boosting Trees, Neural Networks, and Support Vector Machines within systematic fairness optimisation mechanisms integrated with real-time adaptive corrective feedback loops. Empirical testing with data from 15 Chinese universities (a total of 50,000 student records), including in-depth case studies at four representative institutions, showed remarkable performance improvements; the developed framework outperformed traditional methods, achieving 91.3% allocation accuracy compared to 72.4% traditional quota methods and 84.7% baseline machine learning approaches. The fairness analysis reveals significant equity enhancements; reducing regional inequities by over 50% across all geographical regions and achieving a composite fairness score of 0.857. Admission equity was shown to be impacted over all analysed demographic groups. Longitudinal study spanning three years (2019-2022) using historical admission data demonstrates these diverse baseline outcomes and unregulated admission rates converge over time to more equitable outcomes. The examination of stakeholder contentment using a combination of formal interviews and surveys shows over 85% acceptance from government, university, and student stakeholders. Additionally, implementation case studies conducted at four exemplary institutions yield impressive increases in both diversity measures and operational efficiency. This work lays the groundwork for algorithms of equity in the education domain as algorithms of equity are developed for the educational context while offering diagnostic frameworks to the policymakers who wish to use evidence-based strategies for equity advancement in education.