Each city has unique features which impact property prices, and buyers with different value expectations. Valeri is trained using curated data learning sets specific to each city. They include accurate property data, qualitative factors and geographic features not usually applied by other models. These are contiguous data sets across an entire city aligned to geographical coordinates.
PointData carefully cleanses its proprietary data sets for spurious outliers, misentered values and poor or missing features to ensure its machine learning algorithms are trained on good quality data. Data is segmented into various categories to ensure the model is only trained on what it needs for the specific property being valued.
Distinct from other models based on fixed geographies for sample sales, PointData has invented a unique process to dynamically grow the area of influence surrounding each property irrespective of boundary or suburb constraints; we call this the neighbourhood approach. This ensures the model draws on a statistically significant sample of comparable sales (up to 2,000) while keeping the subject property at the centre of the sample area. Valeri only draws on “like for like” properties and sales. When it needs to widen its reach into adjacent suburbs or across council boundaries, it adjusts sample data for local factors.
PointData’s price indexation method at a neighbourhood level allows a more granular level of growth analysis compared to models based on fixed and larger statistical areas (SA2 and SA3). Valeri’s growth indices are calculated separately for a property and its land, in both space (spatially) and time (temporally) to better represent low growth and heterogenous areas. The technology is used to predict future property prices at a neighbourhood level.
Valeri automatically creates a unique mathematical function which describes the land economics for every property.
PointData’s AI-AVMs draw on multiple variables and factors, including property sales, through applying its proprietary data learning sets. In contrast to hedonic and simple AI models that seek best fit and assume property features are homogenous across large statistical areas or suburbs, Valeri uses machine learning to relate “like for like” properties in multi-dimensional space processing all properties within a catchment via Dynamic Neighbourhood Scaling.
PointData applies its unique Land Valuation AI-AVM to assess variables that only affect land price, as opposed to buildings (the capitalised part of a property) which tend to depreciate over time
Many AVM platforms either fix their property price range, apply confidence levels based on data quality or calculate a Forecast Standard Deviation (FSD) based on non-representative samples.
PointData invented an iterative engine that mimics best practice ‘out of sample’ testing. Percent Error (MdAPE) is a class leading 3.3%.
