Smart homes represent examples of cyber-physical environments realizing the paradigm known as ambient intelligence. An information system supporting ambient intelligence takes as input raw sensor measurements and analyzes them to eventually make decisions following final user preferences. Unfortunately, algorithms in this research area are mostly supervised, thus requiring a manual labeling of training instances usually involving final users in annoying and imprecise training sessions. In this paper, we propose a methodology allowing, given a sensor log, to automatically segment human habits by applying a bottom-up discretization strategy to the timestamp attribute of the sensor log. In particular, we show how classical quality measures, computed over Petri nets automatically mined from sensor logs filtered by timestamp, can be used as an heuristic to drive the discretization process, thus providing a likely subdivision of the day in human habits.
Unsupervised Segmentation of Smart Home Logs for Human Habit Discovery / Esposito, Lucia; Leotta, Francesco; Mecella, Massimo; Veneruso, Silvestro. - (2022), pp. 1-8. (Intervento presentato al convegno 18th International Conference on Intelligent Environments, IE 2022 tenutosi a Biarritz; France) [10.1109/IE54923.2022.9826776].
Unsupervised Segmentation of Smart Home Logs for Human Habit Discovery
Esposito, Lucia;Leotta, Francesco
;Mecella, Massimo
;Veneruso, Silvestro
2022
Abstract
Smart homes represent examples of cyber-physical environments realizing the paradigm known as ambient intelligence. An information system supporting ambient intelligence takes as input raw sensor measurements and analyzes them to eventually make decisions following final user preferences. Unfortunately, algorithms in this research area are mostly supervised, thus requiring a manual labeling of training instances usually involving final users in annoying and imprecise training sessions. In this paper, we propose a methodology allowing, given a sensor log, to automatically segment human habits by applying a bottom-up discretization strategy to the timestamp attribute of the sensor log. In particular, we show how classical quality measures, computed over Petri nets automatically mined from sensor logs filtered by timestamp, can be used as an heuristic to drive the discretization process, thus providing a likely subdivision of the day in human habits.I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.