Fraud & Anomaly Detection

Most of what happens in the technical world follows a standard pattern of behaviour. People wake at their usual time, machines are turned on as a fixed time and share prices following a similar pattern of change. Each of these events generates a series of data that can be monitored and used for evaluation purposes.

You wake in a morning, make a hot drink and this generates a spike in electricity usage from the kettle. Your computer is turned on when you arrive in the office, again, creating a rise in electricity usage and a production line runs at a fixed speed, generates a fixed level of heat from the motors and uses a predefined amount of electricity.

However, when these patterns of behaviour change the pattern within the recorded data also changes.

In most cases a delay in boiling the kettle can just mean that you didn’t have to go into work that day or you preferred a soft drink on that occasion. However, if could also signal that an elderly person has fallen during the night and is unable to access the kitchen.

A production line that suddenly generates more heat, makes more noise or uses more electricity may signal the beginning of a catastrophic failure about to take place.

In both these instances any prior warning before or of a change in behaviour taking place can alert a carer to attend or and engineer to inspect the cause of the machine malfunction prior to something disastrous happening.

An automated anomaly detection system would be setup to monitor data as it comes in, during normal operation. Once trained to recognise ‘normal’ it would be set to alert mode and report on anything unusual taking pace.

This form of AI based system is never told what normal is, but learns to recognise patterns unsupervised. Meaning it can be used in a number of systems, as long as a regular stream of data is being produced at regular (time stamped) intervals.