Explainable artificial intelligence for valuation
- Post by: Valentas
- 2021-02-23
- Comments off
Problems in the market
One of the common valuation methods of real estate is the comparative method, which is mainly based on similar transaction selection and adjustment coefficient determination. However, when applying the comparative method in practice several issues are common. Firstly, valuators usually use their own opinion rather than objective reasoning for adjustment coefficient determination. This approach usually is not suitable since the legislation requires providing an explanation of the calculation process. Secondly, only a limited number of valuators apply statistical analysis due to a lack of knowledge. Lastly, there are issues when selecting similar transactions in those regions where the market is not active.
Current valuation process
Currently, valuators when applying the comparative method usually must obtain data of the market, the supply side of the real estate sector, and other related indicators. Then they need to determine criteria based on which to purchase transaction data of real estate, which to use for comparison. During these activities, several issues appear. Firstly, the valuator needs to determine which data sources to use by considering their reliability. In this case, it can be difficult to access all sources and integrate the data for analysis. The selection of similar transactions causes other issues since they need to be purchased in bunches based on the request, this process usually is the most difficult due to costs and time. However, the correct selection of similar transactions is one of the most crucial steps. Lastly, there are difficulties in determining adjustment coefficients. Multiple approaches could be used for this purpose. One of the approaches could be to apply an expert survey, which is important to determine adjustment coefficients that cannot be quantified. Statistical analysis can be used for quantitative data. However, not all valuators have knowledge of how to apply analysis, and not all have a sufficient amount of data. This is a crucial issue, since simple models and a small amount of data cannot represent the trends of the market correctly. However, more complex mathematical approaches are considered as black boxes with explainability that is very limited. Thus, TeGoVa is recommending not to apply complex mathematical models in official valuation reports.
Our solution
Market trend valuation center is developing a prototype of explainable artificial intelligence for expert system development. The solution would integrate multiple data sources, allow automatic selection of similar transactions, and provide a detailed explanation of individual attribute influence on the price of real estate. The solution also will integrate expert survey results, which will allow to validly determine the final price of the real estate object. The development of the prototype is funded by the European Union Funds Investment Operational Program Priority 1 “Promotion of Research, Experimental Development, and Innovation”, Agency for Science, Innovation and Technology, Inostart.