Window-based Mining Sensor Streams

Speaker: Ivy TONG

Abstract

"When stock A and B rises, 95% of the time stock C drops." Rules like this one are useful for a stock investor to determine his action when he sees stock A and B rises. Stock data is one of the examples of sensor stream data. Sensor streams refer to sequences of data continuously generated by a set of sensors. Each reading taken has an associated lifespan of validaty. Since the large volume of stream data makes it infeasible for the whole dataset to fit into memory for mining, window-based algorithms can serve the purpose.

In this talk, I will discuss how we adapt the ideas used in Lossy Counting - a window-based association rule mining algorithm for transactional data stream - to sensor stream scenario. I will also compare different variations of counting and some preliminary experimental results will be presented.