Input Data Measurement Uncertainty Propagation in Artificial Neural Networks

The rapid expansion of artificial intelligence (AI)-based classification and prediction systems has significantly influenced both scientific research and industry. However, ensuring the effectiveness, reliability, and validity of these systems requires rigorous assessment of their measurement uncertainty. A review of current uncertainty evaluation methods and relevant literature reveals a gap in understanding the suitability of artificial neural networks (ANNs) as measurement instruments. This study addresses the key challenges involved by developing an analytical framework that models the internal behavior of the network. The proposed approach enables the propagation of input measurement uncertainty through the network to estimate the corresponding prediction uncertainty, thus providing informed estimates accompanied by measurement uncertainty in accordance with the ISO GUM standard.