Machine Learning: How Facebook Uses it to Detect Fake Accounts

Facebook is filled with fake or suspicious profiles. Machine learning experts believe that it’s a matter of concern for Facebook as well as each one of us. For now, Facebook has taken care of almost 2 billion fake accounts. Wanna learn how? Read below:


Learning Of Blog

  • Introduction
  • The Approach
  • Going Deep
  • Fake friends
  • Conclusion



Facebook – one of the biggest social networks in the world, with almost 2.5 billion monthly active users. But the experts believe that nearly 5% of them are fake. Fraudsters create fake profiles on Facebook to send spam messages, malware, or phishing links. It is a major concern for Facebook’s reputation, and in 2019 it deleted around 2 billion fake accounts, thanks to its new Machine Learning tool developed with the help of machine learning experts. The solution they deployed is called Deep Entity Classification (DEC), a useful machine learning tool to identify fake accounts.


The Approach

Facebook uses a simple approach to classify suspicious accounts – first is misclassified accounts, and the other one is violating accounts.


  • Misclassified accounts – These are accounts that are not appropriately categorized by the users. For example, there’s an account ‘Bruno The Dog.’ Well, technically, this is not a fake, and there’s nothing wrong with it except one thing, they’re not classified correctly. These would have been pages and are converted in the same.



  • Violating accounts – These are profiles that may be fake or a source of widespread spamming, phishing, and other illegal activities. These can also be accounts that violate the company’s terms and conditions.



Facebook uses a combination of a set of rules and machine learning algorithms to prevent fake accounts from causing any harm (preventing the creation of a fake account or preventing its activation). The problem arises when a phony account goes live and is ready to create havoc. This is the point when the machine learning tool DEC comes into play.


Going Deep

DEC is a machine learning tool that offers multi-layered account screening; it is more than a tool that merely monitors account activity. It tracks the social interaction of the account – profile picture, IP address, friends, comments, likes, groups joined, pages followed, etc. And it doesn’t stop there.


It will then perform an in-depth analysis of all the entities connected to the suspected account. It will also monitor the friends, their interactions, and their ages. These measures are enough to give off signals and to categorize it as fake.


The system has learned from a data set that contains both fake and real accounts. In general, machines are trained with two types of data: low precision (labeled by other machine learning algorithms) and high precision (marked by security experts). DEC is trained using millions of samples of lower precision data and then by hundreds of thousands of examples of higher-precision data.


According to a report by The Verge, Facebook has caught more than 6.5 billion suspicious accounts to date, protecting its users from frauds even before they happen. Facebook also claims that DEC has helped them reduce spam by 27% on its website.


Still, there are many active fake accounts on Facebook right now, and as the new tool gets better with time, FB will win the battle eventually.


Fake Friends

Catching every fake account on Facebook is a tricky and never-ending task. There are two significant challenges here:


  • Facebook is enormous for human analysts alone


  • Intelligent machines will still remain machines and cannot be entirely trusted, as there might be a subtle difference between a fake and real account. Not all accounts that look fake are actually fake.



Folks at Facebook believe that they have found the right approach to tackle this problem, and with time, the social network will be much safer.



With Facebook leveraging ML to tackle the problem of fake or suspicious accounts, the future of ML seems to be bright. DEC is an excellent approach by Facebook to ensure that its users are safe from any suspicious activity, and in most cases, Facebook will prevent such attacks even before they occur.


If you too, wish to become a machine learning expert and be a part of a fantastic company like Facebook and work in technologies like AI and ML, check out our machine learning course – Certified Machine Learning Expert.