Streaming Data Classification using Hybrid Classifiers to tackle Stability-Plasticity Dilemma and Concept Drift

摘要

In recent days because of the availability of large number of sensors there is a possibility of generating huge streams of data. Performing data mining tasks such as outlier detection, classification and regression in these streaming data is difficult but it is very much necessary. The main challenges faced by these tasks include online model parameter adaptation, concept drift, the stability-plasticity dilemma, efficient memory models, model benchmarking, adaptive model complexity and meta-parameters. Out of these challenges the adaptation of concept drift with respect to classification task in streaming data and stability-plasticity dilemma has been addressed here. A new framework using offline GMM and incremental GMM has been proposed for adapting to concept drift and to tackle stability-plasticity dilemma. Experiments are carried out using Phasor Measurement Unit data and the results prove that the proposed framework is efficient in terms of adaption of concept drift.