Data Stream Learning with Selective Base Learners of Ensemble Classifiers: Perspectives for Better Information Systems

The exponential growth of data-centric applications necessitates more complex data analysis methodologies capable of modeling highly dynamic and fast data. Recent research works have overcome these challenges by using the concepts like drift learning, heterogeneous classifiers, and so on. The primary challenge with these studies is that relatively few of them consider real-time data generated by data-savvy applications. Traditional methods such as test-then-train and cross validation are not suitable for this type of streaming data. This research study addresses the aforementioned challenge by experimenting with selected machine learning classifiers on the data stream by using a prequential analysis technique (interleaved test-then-train). The experimentation results for the dataset investigated here provide better results in terms of accuracy (about 95%) and in significantly less time.