Application of the random forest algorithm to a model of anemia classification in Peruvian children

Abstract

Introduction: in Peru, in recent years there is a decrease in poverty. However, the prevalence of anemia continues high; it affects 40,00 % of children from six to 35 months of age.

Objective: to identify risk factors or forecasts in the appearance of anemia in Peruvian children.

Methods: a transverse observational study was carried out from the database created for the Demographic and Family Health Survey, by the National Institute of Statistics and Informatics during the years 2015-2019. The population was constituted by 57 410 children from six to 35 months of age, which had hemoglobin exams. 33 independent variables were selected and six procedures were raised with the random forest algorithm. Values of the area indicators under the curve, specificity and sensitivity were obtained.

Results: The procedure that best predicted the presence of anemia, with values for specificity indicators (63,62 %) and sensitivity (65,88 %) more similar, used balanced data with readjustment of the parameters, reduction of the amount of trees and selection of variables.

Conclusions: the five most important independent variables for the model were: child age, conglomerate altitude, number of prenatal visits for pregnancy, moment of the first prenatal control and size of the mother. The study provided scientific evidence about the use of automatic learning algorithms to predict the appearance of anemia based on common risk factors