A deep learning-based CEP rule extraction framework for IoT data


JOURNAL OF SUPERCOMPUTING, vol.77, no.8, pp.8563-8592, 2021 (SCI-Expanded) identifier identifier

  • Publication Type: Article / Article
  • Volume: 77 Issue: 8
  • Publication Date: 2021
  • Doi Number: 10.1007/s11227-020-03603-5
  • Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Academic Search Premier, Applied Science & Technology Source, Compendex, Computer & Applied Sciences, INSPEC, zbMATH
  • Page Numbers: pp.8563-8592
  • Keywords: Complex event processing, Anomaly detection, Deep learning, Internet of things, Air quality
  • Hacettepe University Affiliated: Yes


With the recent developments in Internet of Things (IoT), the number of sensors that generate raw data with high velocity, variety, and volume is tremendously increased. By employing complex event processing (CEP) systems, valuable information can be extracted from raw data and used for further applications. CEP is a stream processing technology that matches atomic events to complex events via predefined rules mostly created by experts. However, especially in heterogeneous IoT environments, it may become extremely hard to define these rules manually due to complex and dynamic nature of data. Defining rules requires accurate knowledge of relevant events by utilizing temporal dependencies and relations among attributes of events. Therefore, there is a need for an automatic rule extraction system which is capable of generating rules automatically from unlabeled IoT data. This paper proposes a generalized framework for automatic CEP rule extraction with the help of deep learning (DL) methods. The proposed framework contains two main phases which are data labeling and automatic rule extraction phases. We compare several DL methods with each other and regression-based methods to evaluate the proposed framework. In addition, several rule mining methods are employed in the automatic rule extraction phase in a comparative manner. The reconstruction error and prediction success rate of our framework are evaluated using an air quality dataset collected from a smart city application. The results indicate that the proposed framework is able to generate meaningful and accurate rules for unlabeled IoT data.