Performance Analysis of an Efficient Ensemble Meta Learning Framework for Streaming Data Analytics

Streaming data is the continuous arrival of data with rapid speed and unbounded in size. Apart from the huge volume and rapid speed of streaming data, it also holds the issue of Concept Drift which leads to drastic drop in prediction accuracy of the model. Therefore, Drift Detection is the major challenge in streaming data which has been attracted by many academicians, industrialists and researchers to provide a novel solution. This paper focusses on a novel solution for the Concept Drift in streaming data, comprises of two phases. In the first phase of the research work, a memory efficient Lasso Regression Hoeffding Tree (LASH Tree) is introduced, as an efficient base learner, with proven experimental studies, and this has been well accepted by the research forum and published in a refereed journal. In the second phase of the proposed research work, an Efficient Ensemble Meta Learning Framework is designed and developed including LASH Tree as base learner, a novel solution to identify and to be adaptable with Concept Drift in streaming data. The experiments are conducted using Solar Power Plant Data Set and it is proved that the proposed framework using LASH Tree consumes less memory and less execution time when compared with the existing ensemble CART and ensemble VFDT algorithms.