The percentage of people with obesity has almost tripled in the last 30 years. It is a true epidemic and, therefore, governments seek to measure its impact. To do this, they use expensive surveys in which, literally, an official goes from house to house with a scale asking people to weigh themselves. Hence the time between measurement and measurement is long.
For example, information is collected every six years, which reduces its impact on public policies. With these problems in mind, Jocelyn Dunstan, a researcher at the Mathematical Modeling Center (CMM) at the University of Chile, looked for a solution in mathematics. And found her.
“We tried to predict the proportion of the obese population. And what we found was that it was possible to do so with a low margin of error,” said Dunstan, who also works at the Center for Medical Informatics and Telemedicine (CMIT).
Together with Chilean and North American researchers from John Hopkins University, he trained three algorithms with two ingredients: information on the obesity rate in 79 countries and databases with sales of 48 product categories in those same nations. For this, he used machine learning, a class of artificial intelligence applications that allows computers to automatically recognize patterns in data after a teaching period.
In Chile, for example, the algorithms gave an obesity rate of 28 percent, only three points below the 31% detected by the government survey.
For the researcher, the results open up great possibilities, since they are data obtained regularly by consultants and governments. This allows to keep a more updated account of the state of obesity in the world, at lower costs.
“The database is small, but of good quality. Thus, one can squeeze it and obtain data much more often than with a survey. Because here we are not talking about big data. No way. Rather, I would talk about small data, “says the researcher.
The team also discovered that three categories
Are the most relevant when predicting obesity: dough and flour, cheeses and carbonated drinks. The selection is relevant, since measuring a limited number of products saves information processing time. This allows to reduce costs and incorporate countries to the sample with greater speed.
In the Chilean case, the study detected an almost 37 percent increase in beverage sales since 2001, with consumption reaching 135.8 liters per person today. In the case of cookies and breads, growth reaches 27.8 percent.
In the investigation, national diets were also portrayed. “You see that Germany and the Netherlands share a lot. All the countries of the Caribbean and Central America are like a big cloud. Colombia, Venezuela, Dominican Republic and Guatemala are very close to each other. One notices the historical, cultural and geographical influence between nations Chile, on the other hand, appears closer to Turkey or Saudi Arabia, just between the Mediterranean and Latin American countries, “he said.