Work detail

Application of machine learning methods for estimating apartment prices in the Czech Republic

Author: Mgr. Jakub Nikodym
Year: 2019 - summer
Leaders: prof. PhDr. Ladislav Krištoufek Ph.D.
Consultants:
Work type: Finance, Financial Markets and Banking
Masters
Language: English
Pages: 77
Awards and prizes: Nomination Deloitte Outstanding Thesis Award
Link:
Abstract: In this thesis, we propose alternative ways to apartments’ mass appraisal. This
work enriches the current literature by combining several techniques of data
extraction and price estimation. We are not aware of any similar work providing
an in-depth overview of the Czech apartment market.
Throughout the empirical analysis, five different methods (OLS, LASSO,
decision tree, random forests, and kNN) are applied to the dataset of 15,848
classifieds. The aim of the study is to find the most accurate method of estimating offering prices, using structured variables as well as data extracted by
text mining. We use various accuracy statistics and graphical analysis to validate our results. Tree-based methods, specifically the random forest algorithm,
results with the highest accuracy in predicting offering prices. Additionally,
text-based variables included in the model cause the reduction of errors on
linear models.
The last part of the analysis covers the main determinants of property
value in Prague and the rest of the Czech Republic. We show that prices in
Prague can be estimated with higher preciseness and with the lower number of
independent variables.

Partners

Deloitte

Sponsors

CRIF
McKinsey
Patria Finance