SOTA Explainable Boosting & Risk-Adjusted Valuation
Learn how Offerlytics trains a State-of-the-Art Explainable Boosting Machine (EBM / GA²M) on 24,898 certified Travis County properties with 29 public record, environmental, and geospatial features to generate 100% auditable, uncertainty-bounded valuations.
1. Explainable Boosting Machine (EBM / GA²M) vs Conventional AVMs
Conventional automated valuation models (AVMs) operate either as rigid linear regressions or opaque black boxes. Offerlytics utilizes a Cyclic Gradient-Boosted Generalized Additive Model (EBM / GA²M) that fits non-linear 1D splines and 2D spatial-feature interactions with real estate monotonic constraints.
Every feature (building permits, TEA school accountability, EPA air quality, USGS elevation, Edwards Aquifer recharge zone, Atlas 14 floodplains, and TCAD CAMA architectural grade) produces an exact Shapley dollar attribution ($\phi_i$) on the audited valuation ledger.
2. 80% Empirical Prediction Range
No valuation model can predict sale prices down to the dollar. PI Value explicitly displays an 80% empirical prediction range (e.g., $798,000 – $869,000) calibrated against held-out market outcomes in Travis County.
When property data is sparse or uncorroborated, the interval automatically widens to reflect increased model uncertainty.
3. Non-Causal Model Drivers
Valuation drivers explain how property-specific signals influenced the mathematical model relative to conventional properties. They describe model correlation, NOT direct physical causality:
Methodology Limitations & Disclaimers
PI Value is a statistical model estimate intended strictly for financial due diligence and decision support. It is NOT a certified real estate appraisal, a tax assessment, a loan underwriting value, or an offer-price recommendation.