A Novel Data-Driven Soft Sensor in Metaverse Provisioning Predictive Credibility Based on Uncertainty Quantification

Seong-Ho Park, Hong Je-Gal, Hyun-Suk Lee

Abstract: Soft sensors are used in various application domains to replace physical sensors. However, the soft sensor’s predictions cannot be trusted as much as the physical sensor’s measurements. Therefore, the evaluation of the credibility of soft sensor predictions is crucial to prepare for the issues caused by inaccurate soft sensor measurements. In this paper, we propose a novel soft sensor architecture that can provide not only predictions but also their predictive credibility via uncertainty quantification. The proposed architecture offers the confidence interval of each prediction that can be used to evaluate the credibility of the soft sensor’s predictions. In our experiments, we build scenarios with different predictive uncertainties and demonstrate the proposed architecture provides credible uncertainty quantification. By using the proposed architecture, we can assess how trustworthy the soft sensor’s predictions are, and thus, effectively manage and respond to varying levels of prediction accuracy.

External IDs:dblp:conf/metacom/ParkJL24