Abstract/Summary

Purpose The aim is to investigate the impact of building condition classifications derived from images on the accuracy of real estate pricing models. It explores whether manual and computer vision-based classifications of three property condition classes can enhance the predictive accuracy of a basic hedonic pricing model for real estate valuation. Additionally, the study sheds light on potential challenges encountered in applying computer vision to real estate valuation. Design/methodology/approach This study explores the influence of building condition classes, derived from images, on a basic real estate hedonic pricing model, using data from online brokerage platforms. The three condition classes are based on a standardized classification and assessed by real estate experts. To examine the potential of computer vision in automating property assessment, the study employs convolutional neural networks (CNNs) to replicate expert condition classifications within the pricing model. Findings The findings of this study indicate that human-based classifications significantly improve the model, while CNNs also enhance accuracy but less effectively. The model’s predictive accuracy decreases when excluding the construction year in CNN-based assessments. CNNs show promise in automating property evaluations with around 60% accuracy for the three predicted classes. Research limitations/implications The research faces limitations due to the subjective nature of manually classifying property conditions and the variability in image quality across real estate platforms. It also notes the challenge of standardizing the use of images for valuation purposes, specifically condition assessment, given the diverse presentation styles. Practical implications It indicates that a shift toward more objective and standardized image assessments could improve the reliability of valuations. This advancement offers practical benefits for real estate professionals and platforms by streamlining appraisal processes and reducing costs. Originality/value Its originality lies in the application of computer vision using real-word data and non-standardized images to automate property evaluations. The value of the research demonstrates the potential for enhancing pricing model accuracy through visual data analysis using data from online brokerage platforms.