Mining of Frequent Patterns from Sensor Data

Speaker: Ivy TONG

Abstract

Mining of association rules is one of the well studied problems in data mining. In the classical association rule mining problem, most of the studies focus on market basket data and there is no temporal correlation in the transaction database. Users are interested in patterns in the form of "The transactions show that there are 80% of customers who purchase product A will buy product B". However, standard algorithms designed to find frequent patterns in market basket data may not be directly adapted to some sensor applications in which the data (states of the sensors) are updated from time to time. Each updated value is valid only until the next update. The time component of the data in these applications should be taken into account during the pattern mining process. An example pattern that users are interested is in the form "In 80% of the time, if sensor A is on, sensor B is on".

In this talk, I will discuss the problem of association rule mining in sensor applications and two approaches to handle the problem will be introduced and evaluated. Some preliminary experimental results will also be presented.

Referred Papers