Enhancing support for machine learning and edge computing on an IoT data marketplace

IoT applications are increasingly employing machine learning (ML) algorithms to manage and control the operational environment autonomously while predicting future actions. To leverage these emerging technologies, the application developers require an enormous amount of data to build models. Data marketplaces enable the IoT application developers to buy data from IoT device owners to train machine learning models. Contemporary data marketplaces only focus on connecting the IoT infrastructure owner (seller) with application developers (buyer) while lacking integrated support for data analytics. Application developers are required to manually create and manage machine learning pipelines by combining edge computing resources with data sources. In this paper, we present an architectural framework to build machine learning pipelines for data marketplaces automatically. Our framework enables application developers (buyers) to leverage the edge computing resources provided by the sellers and compose low-latency IoT applications that incorporate ML-based processing. We present a proof-of-concept implementation on the I 3 data marketplace and outline open challenges in combining machine-learning, AI, and edge computing technologies with data marketplaces.