Kernel Based Estimation of Domain Parameters at IoT Proxy

In this paper we develop lightweight algorithms for monitoring and estimating data lifetime and round trip time at CoAP proxy. We deploy these algorithms in CoAP IoT domain with observe feature with random inter-observation times which can be a consequence of parameterized queries. Algorithms are based on kernel estimation of probability density distributions (pdf). As a result proxy maintains approximate pdfs of these parameters which can be used in congestion control and/or anomaly detection in IoT domain. Results show that estimations with 400-500 samples render satisfactory tradeoff between accuracy and computational complexity even under skewed probability distributions such as exponential distribution.