A Machine Learning Framework for House Price Estimation

Abstract

House prices estimation has been the focus of both commercial and academic researches with various approaches being explored. Depending on the location, size, age, time and other factors, the value of houses may vary. This paper presents a modularized, process oriented, data enabled and machine learning based framework, designed to help the stakeholders within the housing ecosystem to have more realistic estimation of the house prices for decision making. The development of the framework leverages the Design Science Research Methodology (DSRM) and the HM Land Registry Price Paid Data is ingested into the framework as the base transactions data. 1.1 million London based transaction records between January 2011 and December 2020 have been exploited for model design and evaluation. The proposed framework also leverages a range of neighbourhood data including the location of rail stations, supermarkets and bus stops in a cumulative layering approach to explore their possible impact on house prices. Five machine learning algorithms have been exploited and three evaluation metrics have been presented and with a focus on RMSE. Results show that an increase in the variety of parameters enables improved model performance which ultimately will enable decision making. The potential for future work based on this paper can explore the impact of the cumulative introduction of other groups of data on the accuracy of machine learning models designed for the estimation of house prices.