Machine Learning for the Prediction of Anemia in Children Under 5 Years of Age by Analyzing their Nutritional Status Using Data Mining

Abstract

One of the main public health problems is child malnutrition, since it negatively affects the individual throughout his life, limits the development of society and makes it difficult to eradicate poverty. The first objective of this research is to apply data mining techniques for preprocessing, cleaning, reduction and transformation to a data lake that has allowed analyzing anemia in children under 5 years of age, the second objective is to apply Machine Learning algorithms to obtain the best model to predict anemia in children under 5 years of age. The data set was extracted from the open data platform of the government of Peru that corresponds to South Lima, North Lima, East Lima, Central Lima and rural Lima, which collected a total of 138,369 instances and 36 variables of which 30 are categorical and 6 numeric, being an unbalanced data set. In order to obtain the best predictor variables, the Anova F-test and Chi Square filters were used, and it was possible to reduce them to 10 variables, cases were also carried out without considering one of the filters and both filters.To find the best prediction model, the algorithms have been tested: decision tree, logistic regression, K nearest neighbors, random forest and naive bayes. As a result, we show that the best algorithm to predict anemia in children under 5 years of age is the Naive Bayes algorithm with the highest recall of 74%, precision of 43% and accuracy of 70%.