An optimized deep convolutional neural network architecture for concept drifted image classification

Machine Learning (ML) is a branch of Artificial Intelligence, which is continuously evolving to overcome current technological challenges faced by industries. These technological changes are due to modernization in industries for Business Intelligence (BI) i.e., 4th Industrial Revolution. Among the other ML approaches, Image Classification plays a significant role for Business Intelligence and upfront several new challenges in online and non-stationary environment, such as Concept Drift. To overcome the CD issue, one of the fundamental requirements is optimization of classifier. Whereas, Convolutional Neural Network (CNN) is considered best classifier/model for Image Classification. Therefore, the aim of this study is to investigate the optimize architecture for CNN in Concept Drifted environment. This study examines the variety of CNN architectures (CNN1 to CNN4) through different configuration of CNN layers and tuning parameters under certain Concept Drift scenarios. Furthermore, a comparative analysis is performed among these CNN models by monitoring their classification accuracy, loss and computational complexity to validate the optimized CNN model experimentally. In future, proposed Optimize Deep Neural Network architecture will be further investigated for high dimensional Imagery data-streams, for example color and multispectral imagery. © Springer Nature Switzerland AG 2020.