Outlier

We have developed an anomaly detection system using machine learning and adaptive filter methods, to detect abnormal patterns in power traces of computers, servers, and household electric consumption. The motivation behind this system was to detect anomalous power dynamics by a device which is integrated with Google Firebase notification platform to notify users on their mobile devices using push notifications, twitter and/or slack channels, of the detected event. The project was initially developed to notify system admins of organisations of anomalous power traces usage which can be used by hackers to determine the general behaviour and possibly learn the secret encryption keys of the machine. Hence as a test case, we used a power trace from an unmasked version of Advanced Encryption Standard (AES) encryption, also we used a household power consumption dataset from the University of California, Irvine to evaluate our platform. The power traces with regular trends were categorised as normal by our machine learning model. Whereas the anomalous trace/plot clearly indicated outliers in the trace, which immediately alerts the user/admin.

Built With

• android
• firebase
• python
• statistics