Proceedings of the 7th International Conference on Academic Research in Science, Technology and Engineering
Year: 2025
DOI:
[PDF]
Machine Learning-Based Listing Price Estimation with Outlier Detection
Muhammet Ali Kadioglu
ABSTRACT:
Forecasting real estate prices is becoming widespread in many areas such as investment decisions, credit policies, and tax practices. Therefore, the need for accurate and reliable price estimates is increasing daily. Firstly, the peer method is used to estimate the prices of real estate such as housing for sale. This approach is carried out by taking the average unit prices of the real estate with similar features. However, this method is insufficient to reflect sudden changes in market conditions in the estimates. To solve this problem, we developed a model that makes valuations using the prices of villas listed in online channels in the Turkish market with the application we developed. First, we cleaned the dataset from outliers with the limit values we determined based on neighborhood, district, and province. Then, we trained our model with a gradient-boosting framework that uses tree-based learning algorithms.
When we produced a point estimate, we obtained a MAPE value of 18.22%. To meet the business needs and increase the flexibility of the model, we estimated a range—including low, medium, and high values—rather than making a single-point prediction. If the realized value was not within the range we estimated, we accepted the distance of the value to this range as an error. As a result, we reduced the error to 12.20%. The error can be attributed to listings collected from a broad geographic area. Additionally, the error in villa prices arises because they are influenced by land size, but the listings do not provide standardized and validated data for this factor.
keywords: Real Estate Valuation; GBM, IQR, Outlier Detection, Price Estimation