Hedonic Pricing Theory - A Machine Learning Investigatio

The goal of this thesis is an algorithmic investigation of the hedonic pricing theory. The theory suggests that the price of a differentiated commodity is a function of its composite attributes. The objectives of the study are; i) design, development and evaluation of an efficient analytical framework for hedonic pricing analysis ii) design, development and evaluation of a spatio-temporal hedonic regression model with a dimensionality reduction capability iii) design, development and evaluation of a deep learning hedonic predictive regression model and iv) design, development and evaluation of a hedonic pricing software application. My methodology is based on statistical data analytical approach for inference and interpretation, learning algorithm paradigms for sniffing through large piles of data for pattern recognition and knowledge discovery, and software application modeling using model view controller (MVC) architectural design. The novelty of this work lies on our approach in explaining the hedonics (pleasures) of a differentiated commodity from its characteristic features. I argue that there exists a strong and irrefutable evidence of a logical and statistical relationship between the price of a differentiated commodity and its characteristic features. Theoretical and empirical analysis shows; a) an Occam's razor approach of parsimony and plurality to hedonic pricing theory is effective at obtaining set of implicit prices, b) the price of a real estate property is predictable using hedonic theory; deep learning algorithm improved the r-squared value of prediction by 27% as compared to the ridge regression c) grouping a spatio-temporal housing submarkets into smaller clusters has a considerable impact on the dimensionality of the dataset without a significant reduction in the performance of the model; KMLASSO shows a 35.6 % improvement over LASSO and d) house price predictive application software implementable using hedonic theory. The contributions of this thesis include: i) an efficient analytical framework for hedonic pricing analysis, ii) KMLASSO algorithm - a spatio-temporal hedonic regression model with dimensionality reduction capability, iii) a deep learning hedonic predictive regression model, and iv) PREMLS software- A hedonic application software using MVC architecture.