This project aims to develop hybrid non-linear statistical models to predict house prices.
Large banks such as CommBank have models to predict house prices. These models can perform poorly when out-of-sample predictions are required. It is important for these models to accurately predict house prices, regardless of what is presented to them.
This project will explore the use of genetic algorithms (with stochastic search) to predict house prices. Furthermore, as houses may not have publicly available information such as previous sales, or an unusual number of bedrooms, hybrid models will need to be considered. After different models have been produced, statistical studies will be conducted on their performance for out-of-training set scenarios.
The research plan consists of three main steps:
1. Preparation stage: collecting real online data about prices and properties of sold properties and
inspecting their structure.
2. Research and investigate genetic algorithms and their mathematical models. Developing new
modifications for variable selection problems.
3. Investigating hybrid prediction methods based on parametric non-linear models and machine
learning approaches. Studying their properties, including robustness and extrapolate-ability with
applications to property prices prediction problems.
This project involves statistical inference studies and numerical applications. The outcomes of the project are important for economic applications, as well as statistical methodology.