Automated Valuation Model for UK Residential Property

A new level of certainty to property buyers, investors and loan institutions

Automated Valuation Model for UK Residential Property

A new level of certainty to property buyers, investors and loan institutions

Automated Valuation Model for UK Residential Property

A new level of certainty to property buyers, investors and loan institutions

PyTorch, KNN/ANN, Residual Blocks, VAE, ClickHouse, PosgreSQL, Google Cloud, Pandas, NumPy/SciPy

Challenge

Property valuation is a service of huge demand, especially in London, UK, where hundreds of real estate deals are being performed every day. Automated Valuation Models are seen as complementary to traditional appraisals. Experts say that systematic and fast quantitative methods might reduce inaccuracies due to reliance on human judgement.

ENBISYS has developed an AVM system with very high accuracy for property valuation companies in the United Kingdom.

MdAPE
UK: 7%
Greater London: 6%

ENBISYS has developed an AVM system with very high accuracy for property valuation companies in the United Kingdom.

MdAPE
UK: 7%
Greater London: 6%

Approach

We aimed to train our neural network so the valuation is as accurate as possible, considering open data available. The first task was to:

Approach

We aimed to train our neural network so the valuation is as accurate as possible, considering open data available. The first task was to:

Approach

We aimed to train our neural network so the valuation is as accurate as possible, considering open data available. The first task was to:

Mine different types of data from various sources

Clean the data

Set it up for practical use

Mine different types of data from various sources

Clean the data

Set it up for practical use

During this R&D project we developed proprietary methods for Feature Engineering, so we could calculate the influence of the feature on overall AVM accuracy before starting the actual work. During the R&D process, we implemented Deep Learning, the most sophisticated AI technology.

Solution

AVM by ENBISYS is an online platform that allows getting property valuations across the whole UK territory. The Artificial Intelligence algorithms ask for the address, property type, number of rooms and the total area of the property. Based on over 30 parameters, the Customer gets the approximate property price with a certain degree of confidence.

Solution

AVM by ENBISYS is an online platform that allows getting property valuations across the whole UK territory. The Artificial Intelligence algorithms ask for the address, property type, number of rooms and the total area of the property. Based on over 30 parameters, the Customer gets the approximate property price with a certain degree of confidence.

Solution

AVM by ENBISYS is an online platform that allows getting property valuations across the whole UK territory. The Artificial Intelligence algorithms ask for the address, property type, number of rooms and the total area of the property. Based on over 30 parameters, the Customer gets the approximate property price with a certain degree of confidence.

Feature engineering/selection
Neural nets with residual blocks
Prediction confidence estimation
Properties similarity estimation

Neural nets with residual blocks

Feature engineering/selection

Prediction confidence estimation

Feature engineering/selection
Neural nets with residual blocks
Prediction confidence estimation
Properties similarity estimation

Neural nets with residual blocks

Feature engineering/selection

Prediction confidence estimation

Outcome

We collected large amounts of data and came up with our own methods to make almost any data work
for a particular valuation task. Thus, with more data available, it will be possible to increase the AVM accuracy
even further.

Outcome

We collected large amounts of data and came up with our own methods to make almost any data work
for a particular valuation task. Thus, with more data available, it will be possible to increase the AVM accuracy
even further.

Outcome

We collected large amounts of data and came up with our own methods to make almost any data work
for a particular valuation task. Thus, with more data available, it will be possible to increase the AVM accuracy
even further.

BACK TO CASES

Let's discuss your Case!