ADAPTIVE CNN ENSEMBLE TO HANDLE CONCEPT DRIFT IN ONLINE IMAGE CLASSIFICATION

The analysis from the data streams is an essential requirement in the current era of digitalization. However, the critical features of many real-world data streams (imagery streams) such as high-dimensionality, large size, and nonstationary nature lead to concept drift, cause the characteristics of the data streams can change arbitrarily over time. The presence of concept drift renders many classical machine learning approaches unsuitable, hence research community must address this critical issue and contribute towards new adaptive approaches in their place.