It would work by using the data provided by users who answer a series of questions and actions based on their use of the website.
According to the US Patent and Trademark Office website, the algorithm will operate as follows:
An online system uses classifiers to predict the socioeconomic group of users of the online system.
The classifiers use models that are trained using features based on global information about a population of users such as demographic information, device ownership, internet usage, household data, and socioeconomic status.
The global information can be aggregated from market research questionnaires and provided to the online system.
The classifiers input information about a user and output a probability that the user belongs to a given socioeconomic group.
The input information is based on a user profile on the online system associated with the user as well as actions performed by the user on the online system.
Thus, the online system can predict the user's socioeconomic group without using the user's income information.
The online system can generate content for presentation to the user based on the predicted socioeconomic group.
Given that Facebook's targeted advertisements are already scarily accurate, you have to question how will it benefit them to know if you are working or middle class.
Well, the answer is easy: more adverts. Knowing what people's socioeconomic backgrounds are will make it easier for Facebook to filter appropriate information and adverts for a person's timeline based on their history on other sites.
This isn't too dissimilar to what Facebook already does with your information.
But just because they have a patent in place for this type of technology, it doesn't mean they are likely to use it.
As already stated, the patent was filed in July 2016 and it appears that nothing has been done with it since. Furthermore, the algorithm wouldn't be an entirely accurate assessment of your bank balance and will serve as more of an estimate.