engleski

Selection of an Appropriate Prior Distribution in Risk Assessment

The Bayesian approach, which combines prior information about the quantity to be measured, available before the measurement, and additional information obtained from the measurement, is used in risk assessment in metrology. According to the binary decision rule in risk assessment, there are four outputs: the number of accepted and rejected measurements and the number of falsely accepted and falsely rejected measurements. A falsely rejected measurement represents the producer's risk, while a falsely accepted measurement represents the consumer's risk. These four cases in risk assessment lead us to confusion matrix. In this paper, we evaluate the most suitable prior distribution for modelling the risk for roundness deviation of the inner ring of the bearing. This quantity is always positive ; therefore, the choice of prior is limited to those distributions that take only the positive value of the argument. The assessment of the most appropriate distribution was performed by measures derived from confusion matrix and ROC - AUC analysis.

consumer's risk ; producer's risk ; binary decision rule ; confusion matrix ; ROC curve

Best paper award