Please use this identifier to cite or link to this item: https://scholarbank.nus.edu.sg/handle/10635/226593
Title: PREDICTING THE VALUES OF SINGAPORE’S B1 AND B2 INDUSTRIAL PROPERTIES
Authors: AW CHIN KAI
Keywords: Machine Learning
Real Estate
Regression
Industrial Property
Property Valuation
Ensemble Methods
Bagging
Boosting
Stacking
Random Forest
Adaptive Boosting
Gradient Boosting
Issue Date: 2022
Citation: AW CHIN KAI (2022). PREDICTING THE VALUES OF SINGAPORE’S B1 AND B2 INDUSTRIAL PROPERTIES. ScholarBank@NUS Repository.
Abstract: The global market value of all real estate combined is worth more than any other financial market as such real estate valuation will always be imperative and relevant. With the increased application of Machine Learning (ML) Models (MLM), much research has explored its application in the valuation of residential properties. However, MLMs and Ensemble Methods, specifically Stacking Ensembles that perform better and their applications in the appraisal of industrial properties, are not widely researched. Further, filter-based feature selection and Tree-based Ensemble Methods will reduce the subjectivity when selecting factors and adjusting for them during appraisal. The research hypothesizes a set of 66 independent variables that may affect the values of 24 B1 and B2 industrial properties from 2014 to 2020. The data were collected from Ascendas Real Estate Investment Trust’s annual financial reports and building factsheets, government sources, and geographical information system software. They were then processed to remove missing, discrete, and redundant variables to a smaller set of 40 independent variables. The research then uses Ordinary Least Squares (OLS) to test for significance and as a form of feature selection. From OLS, 14 significant independent variables were extracted and used as the final dataset to train and test the performance of Bagging (Random Forest Regressor, Extra Tree Regressor), Boosting (ADA Boosting Regressor, Gradient Boosting Regressor), and Stacking (combination of all previously mentioned models) Ensemble Methods/MLMs in predicting industrial property values. All Ensembles have a testing score of R2 ? 98.8% and mean absolute percentage error (MAPE) ? 5.70% to 13.0%, reflecting their good performance. The good performance of the models comes at a cost. The ML models are sophisticated and require a lot of data that may not be readily available or documented. Further research will need to develop more robust models requiring smaller data sets or multi-modal approaches to handle different data types. It is also of interest to test the models on non-industrial properties.
URI: https://scholarbank.nus.edu.sg/handle/10635/226593
Appears in Collections:Bachelor's Theses

Show full item record
Files in This Item:
File Description SizeFormatAccess SettingsVersion 
Aw Chin Kai DBE_Chin Kai Aw.pdf2.15 MBAdobe PDF

RESTRICTED

NoneLog In

Page view(s)

12
checked on Sep 29, 2022

Download(s)

3
checked on Sep 29, 2022

Google ScholarTM

Check


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.