Data stream mining with limited validation opportunity: Towards instrument failure prediction

Abstract

A data stream mining mechanism for predicting instrument failure, founded on the concept of time series analysis, is presented. The objective is to build a model that can predict instrument failure so that some mitigation can be invoked so as to prevent the failure. The proposed mechanism therefore features the interesting characteristic that there is only a limited opportunity to validate the model. The mechanism is fully described and evaluated using single and multiple attribute scenarios.