Automated Valuation Models (AVMs) use various data points to estimate a property’s value — comparable sales being a significant one.
AVMs like the Xome Value® cannot accurately estimate a property’s value without access to and enhanced processing of comparable sales data. Without this, it is challenging to get a complete picture of current market conditions, which may result in generating a skewed value.
That’s where artificial intelligence (AI) can help. AI automates repetitive tasks in AVMs, such as data cleaning, feature selection, and model training. It can process and analyze large volumes of property data quickly from multiple sources, recognize patterns, and make predictions, all in real time.
There are several ways AI accomplishes this.
Clustering algorithms group similar properties or neighborhoods for comparison.
Clustering algorithms group similar data points by measuring the distance between them and assigning the data to corresponding clusters. The resulting clusters can then be analyzed based on the similarities of the data points within them.
In the world of real estate, these algorithms can group similar properties or neighborhoods based on shared features like location, size, and price. For example, a cluster might include all 3-bedroom homes in a specific zip code with similar lot sizes and recent sale prices. This makes it much easier for AVMs to select the most relevant comparable sales and accurately estimate property valuations with that data.
This can be done with a centroid model like K-Means by partitioning data into k clusters for comparison. This is often after an initial step of using a pre-clustering algorithm like canopy clustering to simplify data for application, which only groups data but doesn’t refine results like K-Means does.
Natural language processing enhances the depth of property comparisons.
Natural language processing (NLP) can enhance the depth of property comparisons by extracting and analyzing unstructured textual data, such as property descriptions. This adds more qualitative data to quantitative data.
For example, NLP identifies and extracts key features from property descriptions that structured data may miss, such as “hardwood floors” and “ocean view,”. This adds more contextual data — data that might otherwise have been missed — resulting in more detailed comparisons.
An NLP technique that does this well is named entity recognition (NER), which uses pre-trained models to detect and classify property features from text, patterns, and context. Custom entities can also be added by training models on real estate specific data.
AI can update models with current market conditions in near real-time.
AI can integrate data in near real-time by continuously importing and processing up-to-date data, such as property listings and sales transactions, from multiple sources.
Tools like Apache Kafka make this possible through near real-time data ingestion, storage and streaming, consumption, and processing. Apache Kafka is a distributed data streaming platform designed to handle high volumes and supports a wide range of data sources and processing frameworks.
How it works is a Kafka cluster is set up with multiple topics for different data streams to read new property listings from a database and publish them to a specific topic, such as “property-listings”. It then updates the machine learning model with the latest data and streams that data to the AVM.
This ensures that the most up-to-date comparable property data is available to generate accurate property valuation estimates
The potential of AI in AVMs and comparable sales
By leveraging the use of AI in AVMs for enhanced data integration, grouping, and processing, it significantly optimizes comparable sales analysis. It has the power to analyze large datasets quickly to find the most relevant properties, enrich the depth of comparisons, and update in near real-time. Because of this, AI provides a more holistic view of the real estate market and property values, even in dynamic market conditions.
As AI continues to evolve and gain popularity, bias is something that may need to be considered when building AVMs. The engineering teams at Xome continually work to improve the Xome Value so that it reflects the most accurate and timely comparable sales data while eliminating bias. They do this by not using any subjective data in the model and reducing regressivity, which is the tendency of a model to perform poorly. Both provide home buyers, sellers, and real estate professionals with clear, data-driven insights to make confident real estate decisions.
Call to action: To learn more about how Xome is leveraging AI and machine learning, read more on the Xome Tech Hub.
