Project

Linking product properties to sensory satiety

Linking product properties to sensory satiety

PhD fellow: Van-Anh Phan

VanAnhPhan1.jpg

van-anh.phan@wur.nl

Supervisors:
Prof. Dr. M.A.J.S. van Boekel (WUR)

Dr. M. Dekker (WUR)

Dr. U. Gazaerek (Unilever Vlaardingen)

Project term:
February 2008 – February 2012

Sponsors:
STW / NOW

Introduction

Food research needs to deal with huge variability and uncertainty due to the nature of raw materials, as well as the lack of human  knowledge in the food - related systems. Bayesian networks (BNs) are a modeling technique where uncertainty and variability are expressed through probability distributions [1]. Bayesian networks models can be used to explain and explore data, and to predict system behaviors. This technique has gained popularity in many fields, such as finance, medical diagnosis, or genetics. Its application, however, has just recently emerged in food-related problems [2].

Aim

The aim of this project is to build a model predicting food intake by physico-chemical properties of foods using BNs technique.

Research

The first phase is to explore Bayesian networks, e.g., to introduce BNs  to the field of food science by  explaining the theoretical background [3] and comparing it with ANOVA models [4] using food examples.

The second phase is to work on combining data from related studies on the impact of aroma, taste and texture on food intake (studied separately) to build a large network.  The difficulty is to get information on variables of interest in one study from the other studies. For example, experiments on the impact of aroma on intake need to report information on taste, and vice versa, in order to support the combination of data from the two studies.

Future research

The project looks also at another approach to the problem of predicting food intake. Rather than combining data of small studies, we will “observe” the variation in sensory properties of foods over a wide range of food products and link this information to intake score. Bayesian networks can help in quantifying these relationships.

In parallel, using BNs in analyzing data from different food areas, such as food quality management and consumer science, is also taken into account. This work is to provide a bigger picture in the potential applications of BNs in food science.

Conclusion

On one hand, challenges when applying BNs in food-related problems are:

-           continuous variables
-           estimation with small data sets
-           get information on a common set of variables from separated studies

On the other hand, this technique allows improved communication of modeller and food experts which is useful for incorporating expert knowledge when modeling and understanding results after modeling and data fitting.

References

1. Heckerman D, Technical report MSR-TR-95 06, Microsoft Research. 1995.
2. van Boekel MAJS, in Bayesian Statistics and Quality Modelling in the Agro-Food Production Chain, 2004, Kluwer Academic Publishers, p. 17-27.
3.  Phan VA, Garczarek U, Dekker M, van Boekel MAJS, in Consumer-driven of food and personal products, Woodhead (Chapter 34)
4.  Phan VA, Weijzen P, Garczarek U, Dekker M, van Boekel MAJS, Translating ANVA models into Bayesian network (in preparation)