How Prediqta works
Prediqta is a gradient-boosting AI price-estimation model that learns from Hungarian property listings that have left the market. Every estimate is indicative and comes with explainable factors.
Model architecture
The core engine is a gradient-boosted decision tree ensemble (XGBoost-class), combined with a geographic embedding layer that encodes location value independently of unit attributes. A meta-model calibrates the ensemble output per segment (new-build vs. resale, urban vs. rural).
Training data
- Recorded sales since 2010, nationwide
- Daily-refreshed listing-portal signal (ingatlan.com and peers)
- KSH (Hungarian Central Statistical Office): district price index, demographics, income
- HÉSZ zoning, transit accessibility, POI density
Training and validation cadence
We retrain the model regularly on fresh listing data and check every training run on a held-out test set.
What it is for, and what it is not
The estimate is the output of a statistical model trained on listing data. We publish no accuracy figure: there is no independent “true value” to measure against, and the deviation varies strongly by segment and market phase.
- It reflects the asking-price level of similar properties, not a guaranteed sale price
- Houses and rural areas have fewer comparable listings, so the estimate can deviate more
- Estimates are less certain for properties needing renovation, unfinished or atypical ones
- In a fast-moving, especially falling market the estimate lags behind prices
- It is not an appraisal: bank, official, or legal purposes require a licensed appraiser
Known limitations
- Atypical layouts can deviate more
- Locations with few listings (rural) trail recent market shifts
- Standalone garden estimates are less certain (fewer comparables)