ML Model Optimizes Sorority Recruitment Quotas with 96.4% Accuracy

A new paper from Omid Khormali and Mia Pesavento introduces a machine learning-guided quota optimization framework for multi-round two-sided matching, specifically applied to sorority recruitment. The system combines a Random Forest classifier trained on historical registration data to generate compatibility scores between potential new members (PNMs) and chapters, integer linear programs for invitation quota optimization in Rounds 2 and 3, and a Deferred Acceptance algorithm for final matching. The framework balances three objectives: fairness, coverage, and efficiency. Using only 282 matched training pairs from five years of de-identified data at a small Midwestern university, the compatibility model achieved a cross-validated ROC-AUC of 0.5822, reflecting the inherent difficulty of predicting social compatibility from pre-recruitment data. However, the system is designed to degrade gracefully when ML signal is weak—fairness and coverage constraints dominate quota allocation under noisy scores.

Despite the modest predictive accuracy, the Deferred Acceptance algorithm replicated actual 2025 recruitment outcomes with a 96.4% individual-level agreement rate and a 100% match rate across 56 PNMs. Optimized quotas also closely aligned with actual coordinator decisions for active chapters. The authors provide an interactive web application for recruitment coordinators to implement the framework. This work demonstrates the viability of data-driven approaches in small-market matching scenarios where data is limited, with broader applications to other constrained two-sided markets such as hiring or college admissions. The paper is available on arXiv (2606.13935).