Aim: To investigate whether eating patterns of specific food groups can be used to predict and classify Mexican adults who have been diagnosed as having obesity, diabetes or both, when compared to those without a diagnosis. Additionally, we aim to show the benefit of data mining techniques in nutritional studies.
Methods: Statistical analysis of self-reported eating patterns based on designated food groups is conducted. Predictive models for health status based on dietary patterns are built using a naïve Bayes classifier.
Results: Clear patterns emerge in the model building where adults are categorised as having obesity, diabetes or both. The model for diabetics showed the greatest degree of predictability, producing sensitivity results 2.4 times higher than the average, using score decile testing. The models for people with obesity and for those with both obesity and diabetes both reported sensitivity doubling the average. Coverage also showed greatest response for the diabetic model, the first decile containing 24% of all diabetics.
Conclusions: Classifier models using dietary habits as inputs succeed in subcategorising Mexican adults based on health status. Diabetics are associated with a very different, and more appropriate dietary pattern (significantly less sugar consumption) for their condition, relative to the non-diagnosed group. Adults with obesity are also associated with a very different, but inappropriate (higher overall consumption), dietary pattern. We hypothesise that obesity, unlike diabetes, is not seen as a sufficiently serious condition, leading to an inadequate response to the diagnosis. Furthermore, data mining techniques can provide new results in nutritional studies.
Keywords: classification; data mining; diabetes; dietary pattern; naïve Bayes; obesity.
© 2018 Dietitians Association of Australia.