Loadstar: Load Shedding in Data Stream Mining

In this demo, we show that intelligent load shedding is essential in achieving optimum results in mining data streams under various resource constraints. The Loadstar system introduces load shedding techniques to classifying multiple data streams of large volume and high speed. Loadstar uses a novel metric known as the quality of decision (QoD) to measure the level of uncertainty in classification. Resources are then allocated to sources where uncertainty is high. To make optimum classification decisions and accurate QoD measurement, Loadstar relies on feature prediction to model the data dropped by the load shedding mechanism. Furthermore, Loadstar is able to adapt to the changing data characteristics in data streams. The system thus offers a nice solution to data mining with resource constraints. 1 Motivation Consider the following scenario. Two cameras A and B set up on highways transmit streams of snapshots to a central server. One snapshot is taken by each camera in each time ...

• Type: Conference
• Year: 2005
• Where: VLDB
• Authors: Yun Chi, Haixun Wang, Philip S. Yu
• Keywords: Data Streams, Database, Load Shedding, Loadstar, VLDB 2005