Beginning Anomaly Detection Using

Figure 8-4.  Service disruptions

Chapter 8   praCtiCal Use Cases of anomaly DeteCtion

---

302

 Banking

In the banking sector, some of the use cases for anomaly detection are to flag abnormally 

high transactions, fraudulent activity, phishing attacks, etc. Credit cards are used by 

almost everyone in the world, and typically every individual has a certain way of using 

their credit card, which is different from everyone else. So there is an implicit profile of 

the individual using the credit card in terms of how they use it, when they use it, why 

they use it, and what did they use it for. If the credit card company has such information 

about the credit card usage of very large number of consumers, it is possible to use 

anomaly detection to detect when a specific credit card transaction may be fraudulent.

Autoencoders are very useful in such an anomaly detection use case. With such a 

case, we can take all the credit card transactions by individual consumers, and capture 

and convert the features into numerical features such that we can assign certain scores 

to every credit card based on various factors along with a kind of indicator as to whether 

the transaction are normal or abnormal. Then, using autoencoders, we can build an 

anomaly detection model that can quickly determine a specific transaction as normal or 

abnormal given everything we know about all the other transactions for a customer. The 

autoencoder does not even need to be extremely complicated; it can be built with just a 

few hidden layers for the encoder and a few hidden layers for the decoder and still have 

pretty decent detection of abnormal activity (otherwise known as fraudulent activity) on 

the credit cards. Figure 

8-5

 is a depiction of credit card fraud.