A Predictive Approach for Monitoring Services in the Internet of Things

Abstract

In Internet of Things (IoT) environments, devices offer monitoring services that would allow tenants to collect real-time data of different metrics through sensors. Values of monitored metrics can go above (or below) certain predefined thresholds, triggering the need to monitor these metrics at a higher or lower frequency since there are limited monitoring resources on the IoT devices. Also, such triggers might require additional metrics to be included or excluded from the monitoring service. An example for this is in a healthcare application, where if the blood pressure increases beyond a certain threshold, it might be necessary to start monitoring the heart beat at a higher frequency. Similarly, the change of the environmental context might necessitate the need to change/update the monitored metrics. For instance, in a smart car application, if an accident is observed on the monitored route, another route might need to be monitored. Whenever a trigger happens, there are optimization-based methods in the literature that calculate the optimal set of metrics to keep/start measuring and their frequencies. However, running these methods takes a considerable amount of time, making the approach, of waiting until the trigger happens and executing the optimization models, impractical. In this paper, we propose a novel system that predicts the next trigger to happen, run the optimization-based methods beforehand, and thus have the results ready before the triggers happen. The prediction is built as a tree structure of the state of the system followed with its predicted child nodes/states, and the children states of these children… etc. Whenever part of that predicted tree actually occurs, one can remove the calculations of the part that did not occur to save storage resources.