Proptech and the proper use of technology for house sales prediction
Using the ATTOM dataset, we extracted data on sales transactions in the USA, loans, and estimated values of property. We developed an optimal prediction model from correlations in the time and status of ownership as well as the time of the year of sales fluctuations.
By Sciforce, software solutions based on science-driven information technologies.
The real estate market has always been tough both for clients looking for a new house and realtors trying to sell properties at the optimal price. In recent years, a bunch of startups has emerged that are aimed at changing the way this market operates.
Their efforts have become known as Proptech — a set of cross-industry technologies changing the way we research, rent, buy, and manage property.
Though in everyday conversation this term is often used to refer to corresponding startups offering technologically innovative products or new business models for the real estate markets, it is, in fact, a much broader notion spanning across software, hardware (such as sensors), materials (think of solar panels) or manufacturing. In this sense, it considers technological and mentality change of the real estate industry, involving our attitudes, movements, and transactions concerning both buildings and cities.
Proptech is only a small part of a wider digital transformation in the property industry, it started to outgrow a niche phenomenon when Venture Capital investments topped 12 BILLION USD in 2017.
It is still a new and growing trend incorporating several verticals: the real estate market (PropTech), smart cities and buildings (IoT), the home building industry (ConTech) and finance (FinTech). Even though ConTech and FinTech are not really PropTech technologies, they are closely connected with the real estate industry.
If we focus on the PropTech sector only, we’ll see the many ways in which startups are using state-of-the-art trends in AI, Blockchain, smart houses, computer vision and virtual reality to push forward the frontier of the real estate industry. And we should not forget that technologies don’t work in isolation: IoT sensors gather data that will be processed using big data analytics and made sense of with the help of AI. Remove one, and the other two will be rendered useless.
Despite the variety of their services or combinations of services, most startups in this field fall into one of two main categories:
- Startups offering support for real estate professionals, such as tools to enhance their services or their productivity;
- Startups supporting clients and proposing to replace real estate professionals.
What about SciForce?
As a company specialized in AI solutions and the IoT, SciForce could not but become a part of the trend. In our recent project, we sided with real estate businesses to make the most of targeted advertising by predicting its impact on house sales.
On the basis of the ATTOM data set, our specialists extracted data on sales transactions in the USA, loans and estimated values of the property. The data processing has given interesting insights into correlations of the time and status of ownership and the time of the year with sales fluctuations, which has become the basis for further experiments with values and thresholds and development of the optimal prediction model.
The results are self-evident:
With the number of ads per year — 1M, the number of houses sold is as follows:
- if we send the equal amount of ads every month — 181 (blue);
- if we send ads using the seasonality feature — 191 (red);
- if we send ads using the seasonality feature for the top 5 months — 218 (yellow).
If we look at the ownership-related features, the difference is even more striking:
Time of ownership
The number of sold houses increases as follows:
- if we send adverts randomly — 281;
- if we send adverts using time of ownership feature — 1422.
We see a similar increase if we take into account the use of the house by the owner.
Owner-occupied (Is the house for rent?)
The number of sold homes:
- if we send adverts randomly — 28;
- if we send adverts using time of ownership feature — 1064.
The Equity position was calculated as AvailableEquity / VM_EstimatedValue, where AvailableEquity was the difference between the current market value and the sum of the current outstanding loan amounts.
Distribution of the equity position
Period of ownership, the actual residence of the owner, and the equity position have become the main features for the prediction model. To avoid the effect of the “black box” and make the decisions traceable and more evident, we have chosen the decision tree classifier. After assigning the optimal weights to the chosen features, we have built a robust prediction model that advises real estate professionals on the best way of sending targeted ads.
The number of ads sent in a month was about 40k.
The number of sold houses increased drastically:
- if we send ads randomly — 146;
- if we send ads using the model prediction — 2408.
As you can see from this pretty small and schematic presentation, technology does play a role and it has the potential to take the real estate industry to a new level, changing the market rules completely and making it less frustrating to potential customers and more manageable to professionals.
Original. Reposted with permission.
Bio: SciForce is a Ukraine-based IT company specialized in the development of software solutions based on science-driven information technologies. We have wide-ranging expertise in many key AI technologies, including Data Mining, Digital Signal Processing, Natural Language Processing, Machine Learning, Image Processing and Computer Vision.
- 6 Industries Warming up to Predictive Analytics and Forecasting
- Airbnb Rental Listings Dataset Mining
- Are you buying an apartment? How to hack competition in the real estate market
|Top Stories Past 30 Days|