Smart Anomaly Detection in IoT

Inspiration

IoT devices generate “Big Data” and there is great need to analyze and act upon that data. The data is available both as historical data and as live stream The techniques for time-series analysis of such data are not as advanced as the techniques for image processing There is a great need to create robust algorithms to address pain-points related to sensor malfunction and operations changes

What it does

Evaluate the Visual Analytics Tool from the EFPF project Demonstrate two new algorithms for anomaly detection Show the implementation in Python with large real time-series datasets

How we built it

The change point determination research was briefly examined and nothing similar to our algorithm was mentioned The ‘stuck sensor” or ‘battery out” sensor situation is common but has not been highlighted adequately in existing research. We provide a ‘simple-minded’ algorithm for that use case.

Challenges we ran into

Restructuring the data

Accomplishments that we're proud of

Both algorithms provided good results

What we learned

Important to address defective sensors and to determine operations changes Our solution works with historical datasets and with live streams The next steps are to improve the algorithms Collaboration with industry partners to get the needed funding will be important