Real Estate Price Prediction Using Machine Learning

Abstract

Background: One of the most significant parts of the global society and economy is Real estate, and
with the rising cost of living coupled with world population increase, the demand for buildings and
properties have also increased. Having a good understanding of prices is important to better make
important and well-informed decisions.

Methods: In this paper, we try to present a machine learning model which can make simple and
accurate predictions from complex data. We also review different ML-based approaches to best
predict the prices of real estate using the most effective algorithms. The methodology involves a set
of comprehensive review of past and existing works and literatures on the field using ML-based
models. Some of the most popular algorithms used for the tasks were Decision trees, Linear
Regression, Lasso, and Random Forest. Additionally, the importance of pre-processing techniques,
feature selection, and tuning of the model to achieve accurate predictions were emphasised.

Results: The study discovers that various algorithms tested performed differently based on the
complexity and size of the dataset, but Linear Regression performed better. It was also observed
that pre-processing and feature selection had a significant impact on the performance of the model.
Using feature engineering, new features were created by combining existing ones to improve the
model’s accuracy. Other implementations such as outlier detection and normalization helped to
enhance the quality and reliability of the prediction. The model outperformed many existing studies
in the field with a 91.5% accuracy score to effectively determine the prices of houses and buildings in
various regions across the UK.

Conclusion: This study provides better insight into the capability and effectiveness of ML-based
approach when predicting the prices of real estate. Prices of buildings across the UK were accurately
predicted based on factors such as square feet, number of rooms, toilet, and location. It also
highlights the benefits of feature selection, pre-processing techniques, and algorithm selection for
achieving predictions that are accurate. It also provides recommendation for future research along
with limitations in the study

Keywords

Real Estate, Price Prediction, Machine Learning, Regression Models, Decision Trees, Random Forests,
Feature Engineering, Data Pre-processing, Feature Selection, Performance Evaluation, Housing
Market, Data Mining, Data Analysis, Data Visualization, Model Interpretability, Regression analysis,
Support vector machines (SVM), Neural networks, Gradient boosting, Root mean square error
(RMSE), Mean absolute error (MAE), Data mining

Research Questions

1. Can machine learning algorithms accurately predict real estate prices based on a range of
data features, including property size, location, and local market conditions?

2. Which algorithms are most effective at predicting the prices, and how do they compare to
traditional statistical models?

3. Can ML models be used to identify undervalued properties and potential investment

5. How effective are the predictions given the constant changes in real estate prices over time,
and how accurate are these predictions?

6. How do the price predictions vary across different geographic regions, and what factors
explain these variations?

7. Can these models be used to provide personalized price predictions for individual properties,
taking into account some unique features such as architecture and design?

8. How do the models perform in comparison to human experts, and what are the implications
for real estate professionals and investors?

Literature review

Introduction:

Real estate refers to any type of property in the form of buildings, lands, or structure that is attached
to land permanently. It can include both residential and commercial properties and natural
resources present on land such as minerals, crops, or water.

The history of real estate in the UK can be traced back to the Middle Ages, when feudal lords owned
vast estates that were worked by serfs. Over time, ownership of land became more diversified, and
the emergence of a market for property developed. Cities grew rapidly during the 18 th and 19th
century industrial revolutions and demand for commercial properties and housing increased. This
ushered in the development of new building materials and techniques like steel and cast iron which
allowed for more efficient and taller buildings to be erected.

In the 20th century, the UK government played an increasingly important role in regulating the real
estate market through various policies, such as rent controls and zoning laws. In the latter half of the
century, there was a surge in private homeownership, which was supported by government
initiatives such as the Right to Buy scheme.

Today, real estate is a major sector of the UK economy, and the market is highly dynamic, with
fluctuations in property values and rental rates affected by factors such as supply and demand,
interest rates, and government policies.

Real Estate in the Context of Machine Learning

Real estate is a significant contributor to the economy of the United Kingdom, and accurate price
prediction is critical for investment decisions. Some of the traditional methods of predicting the
prices involve analysing market trends and historical data. However, machine learning (ML) models
have gained popularity in recent years due to their ability to handle complex data and generate
accurate predictions. In this literature review, we explore the most effective ML-based approaches
to predict real estate prices in the UK and discuss the challenges and limitations of these methods.

Methodology:

The methodology for this literature review involves a comprehensive search of academic databases
such as Scopus, Science Direct, IEEE Xplorer, and Google Scholar for peer-reviewed research papers.
Additionally, we used snowballing techniques to identify relevant papers from the references of the
retrieved papers. We included papers that proposed or evaluated ML-based approaches for
predicting real estate prices in the UK.

The literature review revealed that several ML-based approaches have been proposed for predicting
real estate prices in the UK, each with its own drawback and limitations.

Linear regression models have been widely used for predicting real estate prices in the UK due to
their simplicity and interpretability. In their paper, Wong et al. (2018) proposed a linear regression
model that uses features such as property type, location, and historical prices to predict future
prices. Although it was difficult for them to select an algorithm that worked best for them, they
concluded that choosing an algorithm should be based on various factors such as the knowledge of
the researcher about machine learning, the size of the dataset, and time constraint and with this
they were able to achieve an R-squared value of 0.75, indicating a standard degree of accuracy in
predictions. The study failed to utilise data from a larger geographic region and suggested future
direction considers analysing other property types which are not residential and the incorporation of
more transactional data of properties from a wider geographical location.

Support Vector Machines (SVM) is a powerful machine learning algorithm used for data
classification, prediction, and clustering. SVM was developed based on the principle of structural risk
minimization by Vapnik (1995). Using a hyper-linear separating plane, it creates a classifier and can
perform a non-linear transformation of the input space when necessary Vapnik (1995). SVM can
train classifiers based on different functions and perform regression on continuous and categorical
variables, which is called support vector regression (SVR)

Boser et al. (1992) initially utilized Support Vector Machine (SVM) to recognize handwriting patterns,
and since then it has gained popularity in numerous applications such as text classification, face
detection, and handwriting recognition.

In property research, Li et al. (2009) have utilized support vector regression (SVR) to predict property
prices in China, based on quarterly data spanning from 1998 to 2008. The SVR model used five
features to forecast the property prices as the output variable. Using established assessment
metrics, such as the mean absolute error (MAE), mean absolute percentage error (MAPE), and root
mean squared error (RMSE), their findings demonstrated that the SVR technique is an outstanding
approach for property price prediction.

Rafiei and Adeli (2016) utilized Support Vector Regression (SVR) to decide whether to proceed with a
property development project or terminate it at the outset, depending on the forecast of future
housing prices. In the study, data from 350 condo were employed and these were constructed
between the year 1993 and 2008 in Tehran (Iran). Their model was trained with 26 features,
including estimated construction cost, gross floor area, ZIP code, demographic factors, duration of
construction, housing prices of nearby property developments, lot area, construction duration, and
currency exchange rate. Their findings indicated that SVR is a suitable technique for predicting
housing prices, with a prediction loss (error) as low as 3.6% of the test data. Consequently, the
estimation results offer valuable insights for property developers' decision-making.

Random Forest was an approach proposed by Breiman (2001) by combining regression tree and
classification Breiman et al (1984) and bootstrap aggregation Breiman (1996). It is a regressor or
ensemble classifier which utilises various T decision tree models to derive predictions that are better
in performance. Many trees are created, and a bootstrap technique is used to train the individual
trees generated along with the features which have been randomly selected and partitioned in each
node Pal (2017).

Random Forest was employed in the prediction of housing prices by Wang & Wu (2018) after
collecting more than 27,000 prices of houses from a small county, Arlington in Virginia, USA. Their
model was able to make better prediction than Linear regression based on a set of features such as
age of the building, floor level, floor area, and number of beds. Mohd et al (2019) also utilised
Random Forest to predict the prices of houses in Selangor district of Malaysia. In their study, they
compared results from Decision Tree, Random Forest, Lasso, and Linear Regression and concludes
that Random Forest is the most preferred in terms of accuracy level based on the Root Mean
Squared Error (RMSE).

Koktashev et al (2019) utilised 1,970 transactional records of houses to estimate the value of houses
in the city of Krasnoyarsk. In their study, the number of rooms, floor, total area, type of repair,
parking, type of bathroom, number of elevators and year of construction was taken into
consideration. Their study concludes that Random Forest outperforms Ridge Regression and Linear
Regression which were other algorithms used after the MAE was evaluated.

Boosting algorithm is a technique of numerical optimization developed by Breiman (1998). This

Park & Bae (2015) developed a price prediction model during their property research using ML
algorithms like Ripper, AdaBoost, and Naïve Bayesian after which the accuracy of their classification
was compared. The aim of their study was to help real estate agents or property sellers make
decisions that are rational in property transactions. Their experiment concluded with RIPPER
algorithm outperforming the other models in predicting property prices. Similarly, Satish et al.
(2019) utilised Lasso, gradient boosting, and Linear Regression to predict the prices of houses. The
result of their experiment showed Lasso performing better than the other algorithms. Huang (2019)
employed non-linear and linear ML methods on a sample of 89,412 for Los Angeles housing
transaction records and incorporated features like number of bathrooms, completion year, number
of bedrooms, assessed value, total tax, etc. In his study, he compared the results using methods of
decision tree, random forest, linear regression, SVM, and gradient boosting. All these methods used
had underestimated the Zestimate prediction error and the study concluded that a relative accurate
prediction for housing price was provided by Zestimate.

Another machine learning algorithm used for predicting real estate prices is neural networks. Neural
Networks have shown results that are promising in predicting prices in the real estate industry due
to their ability to handle datasets that are heterogeneous in nature. Xu & Yun. (2021) in their paper
proposed a neural network model that utilises features like location, age of property, and socio-
economic factors to predict housing prices. They used data from a hundred cities across China and
emerged as the first ever study to use such wide coverage. Through examining various
configurations of the algorithm including the delay, hidden neuron, and data splitting ratio, they
were able to develop a neural network that was relatively uncomplicated, with only four delays and
three hidden neurons. This network consistently produced stable results, with an average relative
root mean square error of only 1% across ninety-nine different cities. The study only focused on a
specific parameter for all the cities chosen with a short-term forecast of one month-ahead and this
limitation bore a need for future research which searches for sets of different parameters across
different cities.

Conclusion:

In conclusion, the literature review provides insights into the effectiveness of ML-based approaches
for predicting real estate prices in the UK. It highlights the importance of algorithm selection, feature
selection, and preprocessing techniques for achieving accurate predictions. Additionally, it identifies
the challenges and limitations of these approaches and provides recommendations for future
research to address these challenges. Overall, ML-based approaches have the potential to
significantly improve real estate price prediction in the UK, leading to better investment decisions
and economic growth.

References:

[Link]/10.1080/09599916.2020.1832558

[Link]

[Link]