ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume V-4-2022
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-4-2022, 49–55, 2022
https://doi.org/10.5194/isprs-annals-V-4-2022-49-2022
ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci., V-4-2022, 49–55, 2022
https://doi.org/10.5194/isprs-annals-V-4-2022-49-2022
 
18 May 2022
18 May 2022

STACKING-BASED UNCERTAINTY MODELLING OF STATISTICAL AND MACHINE LEARNING METHODS FOR RESIDENTIAL PROPERTY VALUATION

A. Jafari1, M. R. Delavar2, and A. Stein3 A. Jafari et al.
  • 1Department of GIS, School of Surveying and Geospatial Eng. College of Engineering, University of Tehran, Tehran, Iran
  • 2Centre of Excellence in Geomatic Eng. in Disaster Management, School of Surveying and Geospatial Eng. College of Engineering, University of Tehran, Tehran, Iran
  • 3Faculty of Geo-Information Science and Earth Observation (ITC), University of Twente, The Netherlands

Keywords: Uncertainty Modelling, Support Vector Regression, Weighted K-Nearest Neighbours, Random Forest, Ordinary Least Squares, Stacking, Residential Property Valuation

Abstract. Estimating real estate prices helps to adapt informed policies to regulate the real estate market and assist sellers and buyers to have a fair business. This study aims to estimate the price of residential properties in District 5 of Tehran, Capital of Iran, and model its associated uncertainty. The study implements the Stacking technique to model uncertainties by integrating the outputs of basic models. Basic models must have a good performance for their combinations to have acceptable results. This study employs four statistical and machine learning models as basic models: Random Forest (RF), Ordinary Least Squares (OLS), Weighted K-Nearest Neighbour (WKNN), and Support Vector Regression (SVR) to estimate the price of residential properties. The results show that the integrated output is more accurate for the quadruple combination mode than for any of the binary and triple combinations of the basic models. Comparing the Stacking technique with the Voting technique, it is shown that the Mean Absolute Percentage Error (MAPE) reduces from 10.18% to 9.81%. Hence we conclude that our method performs better than the Voting technique.