Deep Incremental Learning for Big Data Stream Analytics

Internet of Things (IoT) consists of many physical devices that generate data continuously and communicate these data among them. This data is continuously evolving large in volume with time varying nature. Processing these data is critical in all applications including monitoring and control applications. The continuous flow of data and time critical decisions make it infeasible to process the data after storage. Hence, the data must be processed on the fly or online. Online learning or incremental learning algorithms are suitable for processing stream data. Traditional machine learning algorithms are not scalable and so, Deep learning would be of great benefit to analyse steam data. This paper provides a survey on the different approaches used for processing data streams. Concept drift is a significant challenge identified and the paper compares the different works to handle concept drift.