Predictive Void Properties in Social Housing:
Machine Learning
Mitigate the impact of lost revenue, security concerns, additional administration and energy costs from void property across UK Social Housing Stock.
Introduction
Voids can occur due to multiple factors, and social landlords must ensure that no more than 4% of their total stock are in this state at any given time. Once a void property has been identified, it is vital for landlords to relet the asset promptly, fulfilling the housing requirements of a growing number of citizens waiting to be housed.
Often, abandoned properties can remain unoccupied for many months before a landlord is even aware. With many residents in frequent arrears, it’s difficult to know if a tenant is still resident or if they have abandoned their home entirely.
Symptoms of void properties
As we described in our Predictive Maintenance post, it is possible to observe relationships between seemingly unrelated events in data using machine learning (ML) algorithms. Using what are know as Linear Regression Models, we are able to track early symptoms of distant outcomes and train ML models to present an accuracy percentage on their predictions. Typically the earlier in a timeline that a symptom is observed, the weaker the prediction. However as other symptoms present in the data, outcomes like void properties can be identified early and if mitigative measures are taken, avoided completely.
Here are some examples of data feeds that can contribute to ML predictions:
Communication trends & Sentiment Analysis
As most voids occur due to instability of the resident, communication events often present the first indicators. Frequency and channel of communication in isolation can be used to spot trends, however to really leverage the potential of ML, the content of the communications can also be processed using Sentiment Analysis algorithms.
Rent payment trends
Patterns in payment trends also hold a lot of valuable data about the stability of a resident. If payment patterns shift, this can often be an outlier for more severe financial issues like fuel poverty. Maybe a resident is making partial rent payments, or using cash more often than before. ML algorithms can link patterns in the data from properties and tenants that resulted in a void, to train the ML models and increase certainty of predictions.
Facility usage
Where available, facilities and utilities usage can be a key indicator to the human absence in a given property. Door access systems can be considered a facility and produce personal data relating to a resident. Where the use of this data is legally permitted (proven consent from the tenant) usage trends can be used to infer many different outcomes. From irregular usage that link to medical conditions, to the percentage likelihood that the occupant of a dwelling is having fewer overnight stays. This could be due to entering into shared occupancy in another dwelling, or even an indicator that vulnerable residents are spending more time on the streets.
Ingredients to begin?
The only ingredients needed to benefit from this approach is a Bimdl account and some historic data.
The strength and reliability of machine learning outcomes is heavily dependent on the data used to train the algorithm. In essence we can’t predict the future unless we have documented the past. The more training data we feed machine learning algorithm, the more they can learn.
This is exactly why social landlords are a perfect fit to apply this module of Artificial Intelligence, most have been archiving data for decades. Even where the format and structure of their data is abnormal, Bimdl’s trained machine learning models wrangle values to establish an alignment with the training models.
What are you waiting for……your next expensive void property?
Start training your ML model today
You don't need all the data to begin. Simply provide a sample of your historic data and bimdl ML will leverage our broader property data set to strengthen your model.
Submit sample dataSign up here
