An Anomaly Detection Study for the Smart Home Environment

Abstract

Unusual sensor data in smart homes may herald different problems based on sensor errors, security vulnera-bilities, activity and behavior changes. This study focuses on detecting anomalies and unusual situations in 7 different sensor data in a house. For this, a model created with a combination of unsupervised and supervised machine learning algorithms is used. The sensor data are labeled using Isolation Forest which is one of the unsupervised algorithms. Then, the data is trained with the supervised algorithms Decision Tree, Extra Trees, Random Forest and XGBoost classification algorithms. Anomaly decisions are made with an accuracy of over 99 percent.