Time Stamping in the Presence of Latency and Drift

摘要

Online classification learners operating under concept drift can be subject to latency in example arrival at the training base. The impact of such latency on the definition of a time stamp is discussed against the background of the Online Learning Life Cycle (OLLC). Data stream latency is modeled in Example Life-cycle Integrated Simulation Environment (ELISE). Two new algorithms are presented: CDTC versions 1 and 2, in which a specific time stamp protocol is used representing the time of classification. Comparison of these algorithms against previous time stamp learning algorithms CD3 and CD5 is made. A time stamp definition and algorithmic solution is presented for handling latency in data streams and improving classification recovery in such affected domains.