Start Date
25-6-2025 1:00 PM
End Date
25-6-2025 2:45 PM
Description
There is growing interest in using AI within property valuation to transform how valuers and appraisers calibrate mass appraisal models. However, a primary risk of deep learning and tree-based AI methods in mass appraisal is the difficulty of verifying and defending the values they generate. This session explores a contrarian and lesser-known spectrum of AI-based learning techniques—methods that enhance market-based approaches while preserving model interpretability, offering benefits similar to mainstream AI approaches without sacrificing transparency. The session includes concrete quantitative examples of interpretable model calibration and will also touch on emerging hybrid approaches that synthesize latent information such as land rates and depreciation.
Recommended Citation
Wehrli, Joseph, "Interpretable artificial intelligence: Moving beyond least-squares regression without the black box" (2025). Mass Appraisal Valuation Symposium. 7.
https://researchexchange.iaao.org/mavs/mavs2025/sessions/7
Interpretable artificial intelligence: Moving beyond least-squares regression without the black box
There is growing interest in using AI within property valuation to transform how valuers and appraisers calibrate mass appraisal models. However, a primary risk of deep learning and tree-based AI methods in mass appraisal is the difficulty of verifying and defending the values they generate. This session explores a contrarian and lesser-known spectrum of AI-based learning techniques—methods that enhance market-based approaches while preserving model interpretability, offering benefits similar to mainstream AI approaches without sacrificing transparency. The session includes concrete quantitative examples of interpretable model calibration and will also touch on emerging hybrid approaches that synthesize latent information such as land rates and depreciation.