Hier finden Sie wissenschaftliche Publikationen aus den Fraunhofer-Instituten.

CEML: Mixing and moving complex event processing and machine learning to the edge of the network for IoT applications

: José Ángel, Carvajal Soto; Jentsch, Marc; Preuveneers, Davy; Ilie-Zudor, Elisabeth


Association for Computing Machinery -ACM-:
IoT 2016, 6th International Conference on the Internet of Things. Proceedings : Stuttgart, Germany, November 07 - 09, 2016
New York: ACM, 2016
ISBN: 978-1-4503-4814-0
International Conference on the Internet of Things (IoT) <6, 2016, Stuttgart>
European Commission EC
FP7-ICT; 609081; ALMANAC
ALMANAC: Reliable Smart Secure Internet Of Things For Smart Cities
European Commission EC
H2020; 691829; EXCELL
Actions for Excellence in Smart Cyber-Physical Systems applications through exploitation of Big Data in the context of Production Control and Logistics
European Commission EC
FP7-ICT; 614100; IMPRESS
Intelligent System Development Platform for Intelligent and Sustainable Society
Fraunhofer FIT ()
machine learning; complex event processing; stream mining; Internet of Things; edge computing

The Internet of Things (IoT) is a growing field which is expected to generate and collect data everywhere at any time. Highly scalable cloud analytics systems are frequently being used to handle this data explosion. However, the ubiquitous nature of the IoT data imposes new technical and non-technical requirements which are difficult to address with a cloud deployment. To solve these problems, we need a new set of development technologies such as Distributed Data Mining and Ubiquitous Data Mining targeted and optimized towards IoT applications. In this paper, we present the Complex Event Machine Learning framework which proposes a set of tools for automatic distributed machine learning in (near-) real-time, automatic continuous evaluation tools, and automatic rules management for deployment of rules. These features are implemented for a deployment at the edge of the network instead of the cloud. We evaluate and validate our approach with a well-known classification problem.