Weighted Averaging Ensemble Model for Concept Drift Adaptation in Streaming Data

Abstract

In the present world., companies in every industry are overwhelmed by an unprecedented amount of data. The surge of sensor and various network data is a great challenge in deploying a well-fit machine learning (ML) model for data analysis and prediction. These models are subjected to data streams where the statistical properties of the underlying data keep changing. This dynamism results in the decreased performance of the ML models as they fail to generalize to the changes in data distributions. The change in the data distribution is known as concept drift. While there have been many methods introduced to address concept drift., ML models keep suffering from adaptation to the drifts. This paper proposes an error-based weighted averaging ensemble model for adaptation to the concept drift. The model tries to adapt to domains that share the same feature space with varying data distributions. The model is implemented and tested on two real-world datasets and one synthetic dataset. The experimental results illustrate that the proposed ensemble model achieves higher performance compared to the models considered for the study.