Habit monitoring over sensor streams

Julie Soulas (1, 2) , Philippe Lenca (1, 2) , Komate Amphawan (3)

1 Lab-STICC_TB_CID_DECIDE
2 LUSSI - Département Logique des Usages, Sciences sociales et Sciences de l'Information
3 CIL - Computational Innovations Laboratory (Faculty of Informatics - Burapha University)

The use of emerging sensing and communication technologies opens new opportunities for helping elderly or frail people continue living in their home. However, the raw data may be overwhelming for the supervisor. In this poster, we present ongoing work aiming at the discovery of habits from the analysis of the sensor data stream. Habits are searched in the form of episodes, that is to say collections of sensor events that exhibit temporal regularities. These regularities are assessed on combinations of interest measures, such as frequency, periodicity, maximal inter-occurrence gap, length, etc. We propose here structures and methods in order to discover and update over time the k most interesting episodes based on the user preferences on the measures of interest.

Mots clés: Episode mining; Regularity; Periodicity; Event stream; Interest measures; Top-k; Habits

Domaines: Algorithme et structure de données [cs.DS]; Ordinateur et société [cs.CY]

Julie Soulas, Philippe Lenca, Komate Amphawan. Habit monitoring over sensor streams. STATLEARN 2016 : Challenging problems in statistical learning, Apr 2016, Vannes, France. ⟨hal-01417261⟩