Usando regras de associação para a coordenação de objetos autônomos em internet das coisas

Abstract

The efficient coordination of multiple autonomous devices is a central challenge in Internet of Things (IoT) environments. A promising approach is decentralized association rule mining, in which each device locally extracts, from its usage history, the relations indicating which actions tend to occur together, coordinating objects without relying on a central server. However, when such a method is applied to real data, no reference of the true coordinations is available, which prevents directly measuring how much of the real coordination is recovered. This work implements this association rule mining method in a smart residential environment simulator and evaluates it under a controlled ground truth: synthetic personas are modeled by time inhomogeneous Markov chains, with a known coordination structure, and the method's ability to recover the real relations between devices is measured through precision, recall (coverage), and F1. The experiments, carried out over 5 households of 4 residents each, with repetitions and confidence intervals, showed high precision and a coverage that systematically decreases with the size of the coordinated group, an intrinsic limit of one to many coordination by chaining, made measurable by the ground truth evaluation.