Operationalizing predictive analytics for homelessness prevention in Los Angeles
summary
LA County pilots a Homelessness Prevention Unit that uses machine learning on de‑identified CIO InfoHub records from nine agencies to generate risk lists, randomizes those lists before outreach for a planned causal trial, and has called about 400 people and enrolled 45 so far. Models show strong discrimination but drop from AUC 0.895 to 0.823 after stricter housing filters. Outreach is phone‑based with case management and cash assistance; major operational issues are manual data lookup in internal systems and disconnected phones.
AI-written summary · about
chapters (12)
- 0:00Introduction and speaker intro
- 1:26Predicting and preventing homelessness
- 4:56Homelessness policy context in LA
- 8:09Existing prevention services gaps
- 9:54HPU pilot program overview
- 13:42Phase one data challenges
- 17:59Model accuracy adjustments
- 19:25HPU participant profiles
- 21:52Outreach and service delivery
- 23:34Randomized evaluation design
- 26:42Summary and policy implications
- 28:27Q&A session
related talks
35mOn Machine Learning for Prediction and Prioritization in the Allocation of Scarce Societal Resources
34mWhen Prediction Becomes Policy: What Do Risk Models Actually Learn?
47mKeynote: Using Data Science for Social Good: Examples, Opportunities, and Challenges
55mEthical Machine Learning
28mBeyond Prediction Performance: How Modeling Decisions Shape Fairness Outcomes in Statistical...
37mAI for Good- Then and Now
02h 03mStarting Your First Predictive Analytics Project
15mOperationalizing Predictive Analytics in Education