Identifying Food Consumption Patterns among Young Consumers by Unsupervised and Supervised Multivariate Data Analysis
Ulf Hammerling *
Cancer Pharmacology and Computational Medicine, Department of Medical Sciences, Uppsala University, Uppsala Academic Hospital, SE-75185 Uppsala, Sweden.
Eva Freyhult
Cancer Pharmacology and Computational Medicine, Department of Medical Sciences, Bioinformatics Infrastructure for Life Sciences, Science for Life Laboratory, Uppsala University, SE-75185 Uppsala, Sweden.
Anna Edberg
Cancer Pharmacology and Computational Medicine, Department of Medical Sciences, Uppsala University, Uppsala Academic Hospital, SE-75185 Uppsala, Sweden and Råd and Rön, P.O. Box 38001, SE-10064 Stockholm, Sweden.
Salomon Sand
National Food Agency, SE-75126 Uppsala, Sweden.
Sisse Fagt
Department of Nutrition, National Food Institute, Technical University of Denmark, DK-2860 Søborg, Denmark.
Vibeke Kildegaard Knudsen
Department of Nutrition, National Food Institute, Technical University of Denmark, DK-2860 Søborg, Denmark.
Lene Frost Andersen
Department of Nutrition, University of Oslo, NO-0316, Norway.
Anna Karin Lindroos
National Food Agency, SE-75126 Uppsala, Sweden.
Daniel Soeria-Atmadja
Cancer Pharmacology and Computational Medicine, Department of Medical Sciences, Uppsala University, Uppsala Academic Hospital, SE-75185 Uppsala, Sweden and Reveal, P.O. Box 22500, SE-10422 Stockholm, Sweden.
Mats G. Gustafsson
Cancer Pharmacology and Computational Medicine, Department of Medical Sciences, Uppsala University, Uppsala Academic Hospital, SE-75185 Uppsala, Sweden.
*Author to whom correspondence should be addressed.
Abstract
Although computational multivariate data analysis (MDA) already has been employed in the dietary survey area, the results reported are based mainly on classical exploratory (descriptive) techniques. Therefore, data of a Swedish and a Danish dietary survey on young consumers (4 to 5 years of age) were subjected not only to modern exploratory MDA, but also modern predictive MDA that via supervised learning yielded predictive classification models. The exploratory part, also encompassing Swedish 8 or 11-year old Swedish consumers, included new innovative forms of hierarchical clustering and bi-clustering. This resulted in several interesting multi-dimensional dietary patterns (dietary prototypes), including striking difference between those of the age-matched Danish and Swedish children. The predictive MDA disclosed additional multi-dimensional food consumption relationships. For instance, the consumption patterns associated with each of several key foods like bread, milk, potato and sweetened beverages, were found to differ markedly between the Danish and Swedish consumers. In conclusion, the joint application of modern descriptive and predictive MDA to dietary surveys may enable new levels of diet quality evaluation and perhaps also prototype-based toxicology risk assessment.
Keywords: Dietary surveys, young consumers, unsupervised MDA, supervised MDA, dietary prototypes, dietary patterns