Advanced modeling of American household occupancy profiles through data-driven approaches

Document Type

Article

Publication Date

12-15-2025

Abstract

Understanding occupancy patterns plays a crucial role in analyzing residential energy consumption. While national datasets like the American Time Use Survey (ATUS) provide detailed insights into individual activities, they offer a limited understanding of broader household occupancy patterns, which are crucial for energy consumption analysis. This study aims to address this gap by proposing advanced data-driven models to predict household occupancy profiles. To achieve this goal, we integrated individual occupancy patterns derived from ATUS with household information obtained from the American Housing Survey (AHS). We employed data-driven approaches, including artificial neural networks (ANN), random forest (RF), and support vector machine (SVM) techniques, to enhance the accuracy of occupancy status predictions. Results show that the ANN model achieved an overall accuracy of 82 % outperforming RF and SVM. This enabled the creation of detailed household occupancy profiles. These novel American household occupancy profiles highlight how households spend their time throughout the day and offering insights into diverse daily presence patterns. Subsequently, these profiles were transformed into comprehensive household occupancy patterns using the K-means clustering algorithm. This approach not only provides a more precise representation of household occupancy but also can deepen our understanding of the relationship between residential energy consumption and occupancy dynamics.

Publication Source (Journal or Book title)

Journal of Building Engineering

This document is currently not available here.

Share

COinS