Preemptive system failure detection

Fraud, hardware failures, network intrusions and many other issues can be seen as abnormal behaviour within a system.

Data observed from a system when abnormal behaviour is occurring is often significantly different from data observed during normal operating periods. This data can be exploited to detect abnormal behaviour by determining if an observation is an anomaly compared to previous observations. 

We wanted to build a model using machine learning to prevent undesired abnormal behaviour by detecting and notifying someone before it happens by simply using data observed during normal operational periods of a system. Different machine learning techniques were explored for anomaly detection by training and testing them on sensor data that is known to correspond to normal or abnormal behaviour. We present a system that can detect abnormal behaviour - which is useful in preventing real world systems from performing undesirable actions with potentially disastrous consequences.

Theme
Transforming technologies

Booth
TT43

School
Computer Science

Exhibitor
Joshua Chesser

vote for this project: TT43

Back to project list