A Machine Learning-Driven Approach for Proactive Decision Making in Adaptive Architectures

Abstract

Self-adaptation is nowadays considered to be the best solution to dynamically reconfigure a system in the occurrence of deviations from the expected quality of service (QoS) parameters. However, data- and event-driven systems, such as IoT applications, impose new heterogeneity, interoperability, and distribution issues, that make uncertainty on the QoS stability even harder. Typical adaption techniques make use of reactive approaches, an after-the-fact adaptation that starts when the system deviates from the expected QoS parameters. What we envision is instead a proactive approach to anticipate the changes before the event of a QoS deviation. More specifically, we propose IoTArchML, a machine learning-driven approach for decision making in aiding proactive architectural adaptation of IoT system. The approach i) continuously monitors the QoS parameters; ii) predicts, based on historical data, possible deviations from the acceptable QoS parameters; iii) considers a list of possible alternative solutions to prevent the QoS deviation; iv) selects the optimal solution from the list; and v) checks whether the envisioned solution satisfies the overall system QoS requirements. We, therefore, move the focus from self-adaptive architectures to self-learning architectures, enabling the architectures to learn and improve over a period of time.