Adaptive Multilayer Ensemble based Real Time Stream data Classification : Architectural Perspectives and Performance

Recently, there has been a lot of study focused on large-scale data analysis. Data typically arrives continually in the form of data streams in numerous applications of today’s data intensive world. The analytics engines designed to deal with streaming data need to be capable of acting on data -in- motion . Since the conventional techniques for data mining and machine learning were created for static datasets, data streams present unique difficulties. They are unable to effectively analyse data that is expanding at a quick pace and are less suitable to take into account the representative qualities of data streams. The authors through this research viz. A-MERIT-C ( A daptive M ultitiered E nsemble for R eal ( I ntegrated with feature engineering) T ime C lassification), have presented an active learning data stream analysis model through Ensemble Learning Framework ,which is able to learn concepts in near real time streaming data environment. The design of the framework is adaptive to address the dynamism associated with real time data through the classifier pool (i.e. it chooses the best performing classifiers). This design of Adaptive Multitiered Ensemble is composed of prequentially evaluated classifiers rather than traditional hold out evaluation. The distinguishing characteristics of A-MERIT-C give considerable improvements in terms of f1 (87.48), recall (77.79), accuracy (79.27) AUC (88.89) for streaming data analytics as compared to that of similar real time airline FLTRADAR 24 models ; on the other hand, it is able to address the limitations of existing algorithms such as concept evolution , feature drift through incremental learning and feedback .