The use of predictive models to optimize risk of decisions

- Predictive models estimate expected responses and their probability distributions derived from a-priori observations.
- Risk of decision is the expected cost of randomly occurring decision errors. This quantification is far from trivial.
- Decision makers should minimize the cost function, however this may produce counter-intuitive results.
- To develop appropriate cost functions to quantify decision errors is as important as to develop predictors.

The purpose of this paper is to set up a mathematical framework that risk assessors and regulators could use to quantify the “riskiness” of a particular recommendation (choice/decision). The mathematical theory introduced here can be used for decision support systems. We point out that efficient use of predictive models in decision making for food microbiology needs to consider three major points: (1) the uncertainty and variability of the used information based on which the decision is to be made; (2) the validity of the predictive models aiding the assessor; and (3) the cost generated by the difference between the a-priory choice and the a-posteriori outcome.