Investigation of random subspace and random forest methods applied to property valuation data

  • Authors:

  • Tadeusz Lasota;Tomasz Łuczak;Bogdan Trawiński

  • Affiliations:

  • Wrocław University of Environmental and Life Sciences, Dept. of Spatial Management, Wroclaw, Poland;Wrocław University of Technology, Institute of Informatics, Wrocław, Poland;Wrocław University of Technology, Institute of Informatics, Wrocław, Poland

  • Venue:
  • ICCCI'11 Proceedings of the Third international conference on Computational collective intelligence: technologies and applications - Volume Part I
  • Year:
  • 2011

Quantified Score

Hi-index 0.00

Visualization

Abstract

The experiments aimed to compare the performance of random subspace and random forest models with bagging ensembles and single models in respect of its predictive accuracy were conducted using two popular algorithms M5 tree and multilayer perceptron. All tests were carried out in the WEKA data mining system within the framework of 10-fold cross-validation and repeated holdout splits. A comprehensive real-world cadastral dataset including over 5200 samples and recorded during 11 years served as basis for benchmarking the methods. The overall results of our investigation were as follows. The random forest turned out to be superior to other tested methods, the bagging approach outperformed the random subspace method, single models provided worse prediction accuracy than any other ensemble technique.