Understanding Housing and Homelessness System Access by Linking Administrative Data

2025-05-13

AI Summary

This paper addresses privacy-sensitive entity linkage across 235,000 inter-agency administrative records in a major North American city’s Housing and Homeless Services Continuum (HHSC). We propose a privacy-preserving linkage framework integrating differential privacy, cryptographic hashing, and multi-method pairwise entity resolution with clustering-based post-processing. Departing from conventional ML-centric evaluation (e.g., F1-score), we introduce HHSC-domain metrics—including shelter stay frequency, service duration, and number of accessed agencies—as primary evaluation criteria, revealing substantial misalignment between standard ML metrics and real-world system impact. Experimental results demonstrate that our approach significantly improves accuracy in reconstructing individual service trajectories while rigorously preserving privacy—enabling more precise interventions and evidence-based resource allocation within the HHSC.

Technology Category

• Machine Learning: Privacy
• Humans and AI: Human-in-the-loop Machine Learning
• Natural Language Processing: Ethics — Bias, Fairness, Transparency & Privacy

Application Category

• Security and Privacy: Data transparency and provenance
• User Modeling, Personalization and Recommendation: User privacy protection in personalized systems
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

This paper uses privacy preserving methods to link over 235,000 records in the housing and homelessness system of care (HHSC) of a major North American city. Several machine learning pairwise linkage and two clustering algorithms are evaluated for merging the profiles for latent individuals in the data. Importantly, these methods are evaluated using both traditional machine learning metrics and HHSC system use metrics generated using the linked data. The results demonstrate that privacy preserving linkage methods are an effective and practical method for understanding how a single person interacts with multiple agencies across an HHSC. They also show that performance differences between linkage techniques are amplified when evaluated using HHSC domain specific metrics like number of emergency homeless shelter stays, length of time interacting with an HHSC and number of emergency shelters visited per person.

Problem

• Linking housing and homelessness data to track individual interactions
• Evaluating privacy-preserving linkage methods for multi-agency data
• Assessing linkage techniques using domain-specific homelessness metrics

Innovation

• Privacy preserving methods link 235,000+ records
• Machine learning evaluates pairwise linkage and clustering
• Domain specific metrics assess HHSC system interactions