Dataset Evaluating Human–Machine Collaboration through a Comparative Analysis of Experts, Machine Learning, and Hybrid Approaches in Real Estate Valuation

Dataset description

The dataset was collected to support controlled experiments evaluating the predictive performance and efficiency of different residential property valuation approaches. Specifically, it enables a direct comparison between an AI-based price prediction model, human real estate experts, and a hybrid human–machine approach.

The underlying machine-learning model was trained on 21,736 apartment transactions from Vienna covering the period 2018–2022. This transaction data, originally compiled and processed for the study “Location, Location, Location: The Power of Neighborhoods for Apartment Price Predictions Based on Transaction Data” published in the ISPRS International Journal of Geo-Information, served as the empirical basis for model development.

Building on this foundation, the present dataset focuses on the experimental evaluation phase rather than transfer learning. It contains expert assessments of newly built apartments sold in Vienna in 2023, collected under three experimental conditions: (i) limited information, (ii) state-of-the-art expert valuation methods, and (iii) collaboration between experts and the ML model. The dataset further includes the corresponding model predictions and ground-truth transaction prices, enabling a systematic comparison of predictive accuracy and task efficiency across valuation strategies.

This dataset was used to analyze the relative strengths of standalone ML models, human expertise, and hybrid human–AI collaboration in residential price prediction, with particular emphasis on accuracy, robustness, and time efficiency.

Context and methodology

The data set was created to predict of apartment prices 1 to 7 years into the future

The data set was used to test of transfer learning capabilities

Data collected from apartment ownership transactions, enriched by contextual information from OpenStreetMap. The features added were selected based on experience with valuation and discussions on potentially relevant factors

All personal data were removed from the expert survey and the transaction data

Technical details

csv-File with raw data; further explanation in ReadMe.txt

Python-script to analyse the data: PSFL

Licenses

Data: CC by 4.0 International

Code: PSFL 2.0