Abstract
In the ever-evolving world of real estate, owners are constantly looking for new ways to maximize their profits. This research explores the impact of 3D models on real estate sales performance. By analyzing data from the US housing market (Zillow), this study examines the historical performance of real estate listings that used 3D models in their selling efforts. This study presents a comprehensive evaluation of the machine learning models employed to predict the sales performance of real estate properties, focusing on the influence of 3D models. The performance of the price prediction models, including GradientBoostingRegressor, RandomForestRegressor, SupportVectorRegressor, and Multilayer Perceptron, was extensively analyzed and optimized using different parameter configurations. The ensemble approach, particularly the GradientBoostingRegressor, exhibited superior performance in terms of root mean square error (RMSE) and mean absolute error (MAE).Similarly, time performance on market prediction models was evaluated, with the ensemble approach demonstrating consistent performance. Benchmarking against Zillow's estimates (Zestimates) validated the selected models' accuracy, showcasing low average errors of less than 1%. Prediction interval analysis highlighted the models' ability to provide reliable price projections, with a significant percentage of actual prices falling within the predicted intervals. Sensitivity analysis revealed that including 3D models significantly influenced price predictions, resulting in an average increase in expected selling prices by nearly 7%. Overall, the machine learning models employed in this study exhibited strong predictive capabilities for estimating property prices. Additionally, this research expands the existing literature on machine learning in the real estate industry, contributing to the field's knowledge and understanding of the effectiveness of virtual worlds and 3D modeling in real-world assets and transactions. Ultimately, this research aims to provide significant evidence that using 3D models in property listings is beneficial to real estate owners, offering a competitive edge in today's dynamic market.
| Date of Award | Aug 2023 |
|---|---|
| Original language | American English |
| Supervisor | Davor Svetinovic (Supervisor) |
Keywords
- Generative Adversarial Networks (GANs)
- Natural language processing (NLP)
- Artificial Intelligence (AI)
- Automated Valuation Models (AVMs)
- Root Mean Square Error (RMSE)
- Mean Absolute Error (MAE)
- Principal Component Analysis (PCA)
- Explained Variance Regression Score (Evar)
- Coefficient of Determination (R2)
- Zillow
- Zestimates
- Virtual Assets
- Three Dimensional Model (3D Models)
Cite this
- Standard