Housing price predictions using deep learning techniques

Abstract

There is always a demand for housing each year due to the rising population and immigration. New households wanting to purchase and own their first flat would need to consider carefully what type of property they want, as it will put a strain on the annual income of an average household. House sellers would want to consider their house price valuation before deciding to sell the house. Therefore, having a housing price prediction model can assist these parties in understanding the property market in Singapore and also help policymakers in the establishment of housing policies. This report explored the use of machine learning algorithms such as neural networks to predict the Housing and Development Board resale flats in Singapore. It analyzed the past transacted housing data of 48,053 HDB resale flats across 23 towns and 3 estates in Singapore obtained from an online portal managed by the Government Technology Agency of Singapore. Cross-validation was applied to K-Nearest Neighbors, General Regression Neural Networks and Artificial Neural Networks for model tuning before fitting the test set. The experiments show the General Regression Neural Networks model, based on the lowest error rate, was able to outperform the other individual models. Following that, the ensemble learning blending technique was proposed to further improve the housing price prediction model. The experiment demonstrated that by strategically combining multiple individual models, the blending ensemble learning technique could improve the results. The conclusion made is that machine learning and deep learning such as neural networks provide a promising alternative in property valuation and more research should be carried out on Singapore’s property market.