P26 - Robust Indoor Air Quality Monitoring: Out-of-Distribution Detection using Ensemble Neural Networks
- Event
- 17. Dresdner Sensor-Symposium 2024
2024-11-25 - 2024-11-27
Dresden - Band
- Poster
- Chapter
- 14. Smart Sensors/Künstliche Intelligenz in der Sensorik
- Author(s)
- P. Goodarzi, D. Arendes, C. Bur, A. Schuetze - Saarland University, Saarbrücken/D
- Pages
- 160 - 164
- DOI
- 10.5162/17dss2024/P26
- ISBN
- 978-3-910600-04-1
- Price
- free
Abstract
Data-driven indoor air quality (IAQ) monitoring systems have demonstrated strong performance; however, detecting out-of-range data is essential for reliable monitoring. This study proposes an out-of-distribution (OOD) detection method to identify out-of-range conditions and temporal drift in real-time applications. Our approach utilizes an ensemble of convolutional neural networks (CNNs) optimized via Bayesian hyperparameter tuning. The method achieved robust results, with an area under the receiver operating characteristic curve (AUC) of 93% for out-of-range gas detection and AUCs of 95% and 99% for identifying temporal drift at six and ten weeks postcalibration, respectively. Integrating this method into real-time IAQ monitoring systems enhances model reliability under real-world conditions.