Predicting Real Estate Price Variations using Machine Learning and Google Trends

Abstract

The goal of this paper is to create a modern model via the use of machine learning (such as support vector regression, regression tree and neural networks) and google trends to predict real estate price variations. The model should achieve significant predictive capabilities in monthly variations and should be both interpretable and not overly complex. There is major interest in being able to predict real estate prices and many articles have been published on the subject. Most traditional models use economic data which are usually published quarterly or annually and thus are not very efficient for short term predicting. As an investor, real estate has always been an asset class of interest for its performance, diversifying effect on a portfolio and its interest to a short or long term investor. The interest in the subject goes beyond investors as it is one of the most important costs for a regular family. These models will use as inputs various variables that effect either directly or indirectly prices in real estate. We will focus on the Miami metropolitan area or the Miami-Fort Lauderdale-Pompano Beach area. The US market was chosen because it provides the best access to reliable and consistent data. Our model will also focus on predicting single family house prices which are very popular in the US.