Development of a machine learning web application for real estate price prediction

Abstract

The objective of this thesis was to create and assess a machine learning based web application that can predict, in real time, the price of a house based on property features provided by the user. The thesis was interested in developing a practical, simple to use tool that utilizes predictive modelling to provide an estimate of a residential property's value. The main research questions were centred on identifying the most influential factors impacting house prices and the accuracy of machine learning predictions compared to traditional appraisal methods. This thesis is practice oriented. First, it sets out important concepts around housing market mechanisms, regression modelling, and web-based application design. The project then progressed through the steps of acquiring a suitable housing dataset from real world, cleaning and preprocessing the data, engineering features, training an assortment of regression models, and building the web-based interface in Streamlit that integrating the best performing regression model. The main method of research was development based, with some quantitative assessment presented. The machine learning model was created and tested using scikit-learn and reported using a selection of standard metrics such as R² value, MAE, and RMSE. Furthermore, the app had batch prediction and visualization functionality to enhance usability and better understanding. The research concluded that the Random Forest Regressor model can model house prices with accuracy, and when trained on engineering features, the application allowed for single prediction and batch predictions, and data visualization including predicted price distributions. The analysis showed that machine learning can enhance the accuracy of property valuations and the accessibility of these tools. Based on the results of this analysis, it is recommended that further development of the application take place for commercial use by real estate professionals or by individual homeowners looking for quick data driven property valuations.