Anemia prediction among Rwandan children aged 6 to 59 months using machine learning techniques

Abstract

Anemia, especially in children aged 6 to 59 months, remains to be one of the problems affecting public health in Rwanda. Anemia at a young age has major effects on children's developmental delays and impaired cognitive development. Therefore, there is a need for an accurate, reliable, and effective system or model that provides the likelihood of anemia given factors and assists in diagnosis at an early stage of anemia in time for proper treatment. This study aimed at building the best predictive model of anemia among Rwandan children aged 6 to 59 months by using the risk factors through machine learning techniques. The Data used in this research were mined from RDHS 2019-2020 as the nationwide cross-sectional survey, whereby the KR and PR datasets were merged to obtain the necessary information. The participants in this study were 3525 children 6-59 months old whose mothers were interviewed. Several models were developed based on the stratified 8-fold cross-validation using five different machine learning algorithms such as Random Forest, XGBoost classifier, Support Vector Machine, Logistic Regression, and Naïve Bayes for the anemia prediction based on significantly associated factors. After an exhaustive evaluation of the predictive models through various measures including accuracy, sensitivity, specificity, F1 score, and AUC then, an outstanding model was found. Based on the performance measures of different models, the best model was generated by the XGBoost classifier, which was chosen as the best classification with an accuracy of 85.21%, sensitivity of 98.71%, specificity of 71.70%, F1 score of 86.97%, and the area under the curve of 0.98. After modeling, the results indicated that the age of a child, child malaria infections, maternal anemia, children breastfeeding status, deworming and vitamin A in the last 6 months for a child, the residence of a child, and the household wealth index were the most contributing predictive features of anemia status. Overall, this study has proved that machine learning techniques can be considered in building an appropriate model, which can be helpful in the early stage of anemia detection among children and figure out the most important predictive factors, which can assist in controlling and preventing anemia in Rwandan children 6-59 months old.