Phan VA 1, Weijzen P 2, Garczarek U 3, Dekker M 1, van Boekel MAJS 1

1) Wageningen University, PDQ, Wageningen, the Netherlands

2) FrieslandCampina Corporate Research, Deventer, The Netherlands

3) Unilever Food and Health Research Institute, Vlaardingen, the Netherlands

This presentation is part of the workshop "Structural equation and Path Modelling: a useful tool in sensory and consumer science" (Chair persons: Dr Mario Mazzocch and Dr Hal MacFie)

Food research is highly complex and reflects large variability and uncertainty.Bayesian networks, also referred to as Bayesian belief networks, beliefnetworks, Bayes nets, or causal probabilistic networks are a modern dataanalysis tool that can handle variability and uncertainty using probabilitydistributions. These techniques can be used for explanation, exploration ofinformation and prediction of system behaviors and for decision making underuncertain conditions [1]. Despite the popularity of Bayesian networks in variousfields, such as finance, medical diagnosis, robotics, genetics, its applications infood-related problems have only recently emerged [2].

Ideas and techniques of Bayesian networks are going to be introduced throughfood examples. Some published Bayesian networks are reviewed, e.g., a modelquantifying the risk of consumers to C. botulinum neurotoxin during themanufacture and storage of processed potato [3] and a model relating sensoryfeatures to consumer preference in food design [4]. The construction andadvantages of Bayesian networks are demonstrated and compared with theoutcomes of classical statistical analysis based on a sensory specific satietydataset [5]. Most current Bayesian network algorithms require discretevariables, while continuous variables are common for food applications. Stillthese techniques are valuable in interpreting interactions, reasoning in complexsystems, and facilitating communication between statisticians and domainexperts. Modeling with Bayesian networks particularly enable the use of expertknowledge as well as the combination of data from different studies.

1. Heckerman, D., Technical report MSR-TR-95-06, Microsoft Research.1995.

2. van Boekel, M.A.J.S., in Bayesian Statistics and Quality Modelling in the Agro-Food Production Chain, 2004, Kluwer Academic Publishers, p. 17-27.

3. Barker, G.C., et al., Int J Food Microbiol, 2005. 100(1-3): p. 345-57.

4. Corney, D.P.A. In Parmee, I. (Ed.), Evolutionary Design a