Real-time and Energy-efficient Detection of Anomalies in Sensor Networks

The key idea underlying this doctoral research work is to devise a model which detects anomalous activities learning continuously from past activity and current changes. The model should be able to learn in real-time and incremental fashion from the incoming streams of individual data. As this stream could be very large, it should keep a bounded memory footprint to avoid storage issues. Furthermore, in wireless sensor networks case not only information is collected from the nodes but intelligence is also shifted from the central system downwards to these nodes. In such manner, the communication bandwidth is reduced which will lower the over all energy consumption that is one of the bottleneck in such systems.