Comparación entre los modelos de regresión logística y redes neuronales artificiales para determinar los factores maternos que influyen en el peso del bebé al nacer en instituto ecuatoriano de seguridad social de la ciudad de riobamba en el año 2020 – 2021.

Abstract

Low birth weight gas become a worldwide health problem, negatively affecting the short- and long term development of the baby. The main objective of this study was to determine the maternal factors that influence newborn birth weight at IESS-Riobamba in 2020-2021. The database was obtained from the statistics area, with a total of 273 women. For the processing of the models, the R software was used to code the qualitative variables and subsequently create the models of the 2 techniques. The descriptive statistical analysis showed that the minimum age of the mothers was 15 years and the maximum age was 46 years; it was determined that the most common diseases during pregnancy were preeclampsia, gestational diabetes, among others. Finally, it was found that 35.5% of the newborns were underweight while 64.5% were of adequate weight. In the logistic regression model, 7 significant variables were obtained: toxic habits, weeks of gestation, maternal weight, parity, miscarriages, pregnancy diagnosis and BMI, while in the neural network model a model with 2 hidden layers was obtained with 6 and 3 neurons respectively. Finally, in the ROC curves, an area under the curve (AUC) of 91.8% was obtained in logistic regression and 85.9% in artificial neural networks, concluding that the logistic regression model is the best model for determining factors. It is recommended that public and private health institutions interested in this research disseminate this information to pregnant mothers so that they can take preventive measures to prevent their babies from being born with low birth weight.