A Novel Load Shedding Framework LS-LG for Similarity Querying on Data Stream

It is important to obtain effective feature values of data stream and forecast them in overload system for mining data stream, because data streams are often bursty and data characteristic vary over time. In this paper, we introduce linear predictive coding (LPC) technology to obtain feature values using fewer coefficients. Generalized autoregressive conditional heteroscedastic (GARCH) -generalized regression neural network (GARCH-GRNN) model is used to forecast the feature values of which the data streams are shed, and we perform similarity search using these forecasting values. A load shedding framework based on LPC and GARCH-GRNN (LS-LG) for similarity search on data stream is constructed to achieve minimized mining loss. Experimental results indicate that LS-LG is an effective method in improving query quality when the system is under overload situation.