New STRATA model improves race inference for fair lending applications

Researchers have developed STRATA, a new model designed to infer race and ethnicity for fair lending and housing equity applications. Unlike previous methods like BISG, STRATA integrates name sequences with census tract geolocation using Bidirectional LSTM networks and XGBoost, significantly reducing socioeconomic bias. The model achieved 88.7% accuracy on a voter registration dataset and 84.8% on a national loan dataset, demonstrating its effectiveness and generalizability. The developers emphasize that STRATA is intended for aggregate analysis, not individual decision-making. AI

IMPACT Enhances tools for detecting and mitigating bias in financial and housing sectors.

RANK_REASON Academic paper detailing a new model and its performance metrics. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Bayesian Improved Surname Geocoding
• Community Reinvestment Act
• New York City
• STRATA
• Terry Leitch
• XGBoost

• paper
• policy

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →