You Are What You Use: Usage-based Profiling in IoT Environments

TL;DR

This paper presents a novel ensemble clustering approach for habit extraction in smart homes, addressing challenges in activity timing and providing confidence scores for habit occurrence.

Contribution

It introduces an innovative ensemble of unsupervised clustering techniques with new metrics for more accurate habit identification in IoT environments.

Findings

Effective habit extraction using multiple clustering algorithms.

Introduction of a noise metric for improved habit detection.

Habits are associated with time intervals and confidence scores.

Abstract

Habit extraction is essential to automate services and provide appliance usage insights in the smart home environment. However, habit extraction comes with plenty of challenges in viewing typical start and end times for particular activities. This paper introduces a novel way of identifying habits using an ensemble of unsupervised clustering techniques. We use different clustering algorithms to extract habits based on how static or dynamic they are. Silhouette coefficients and a novel noise metric are utilized to extract habits appropriately. Furthermore, we associate the extracted habits with time intervals and a confidence score to denote how confident we are that a habit is likely to occur at that time.