How Prediqta NYC works
Prediqta NYC is a submarket-tuned AI valuation engine for New York residential real estate — same gradient-boosting core as the Hungarian product, calibrated on NYC transaction data with submarket-level confidence bands.
Model architecture
Gradient-boosted decision tree ensemble with a submarket embedding layer. Submarkets are clustered at finer-than-borough granularity to capture micro-market premiums. A calibration head adjusts the confidence band per submarket based on transaction density.
Training data
- NYC ACRIS recorded transactions
- Public listing-portal signal (StreetEasy, Zillow class), refreshed weekly
- Building-level attributes (year built, building type, elevator, doorman) where available
- NYC submarket density and amenity data
Training and validation cadence
Retrained monthly on the latest ACRIS feed; listing signal updates weekly. Hold-out validation per submarket. Model versions are pinnable for audit cycles.
Accuracy and confidence
Accuracy varies by submarket density. Manhattan and dense Brooklyn submarkets are tightest; outer-borough estimates ship with wider confidence bands. Every valuation includes a per-submarket confidence interval so downstream users know the model's certainty.
Known limitations
- Co-op valuations are more uncertain than condos
- Outer-borough small-building rentals have sparse comparables
- The office-to-residential calculator is a first-pass screen — not a replacement for an architect's feasibility study