An Ensemble Approach to Improve the Performance of Real Time Data Stream Classification.

Joshi, Dhara; Shukla, Madhu

In the era of the Internet of Things (IoT), data stream mining has gained importance to make accurate and profitable decisions. Various techniques are used to gain insight into data streams, including classification, clustering, pattern mining, etc. Data are subject to changes over time. When this happens, predictive models that assume a static link between input and output variables may perform poorly or even degrade, which is called concept drift. This study proposes an ensemble architecture designed to improve performance and effectively detect concept drift in stream data classification. Using an ensemble approach, the proposed architecture incorporates three classifiers to improve accuracy and robustness against concept drift. The proposed architecture provides drift detection that ensures the model's continued performance by enabling it to be quickly modified to changing data distributions. Through comprehensive testing, the performance of the proposed algorithm was compared with existing methods, and the results demonstrate its superiority in terms of classification accuracy, precision, and recall and drift detection capabilities.

INTERNET of things; DATA mining; PREDICTION models; CLASSIFICATION; ALGORITHMS

Engineering, Technology & Applied Science Research, 2024, Vol 14, Issue 6, p17749

2241-4487

Academic Journal

10.48084/etasr.8563