10th International Congress on Information and Communication Technology

Authors - Byron Albuja-Sanchez, Jeniffer Flores-Toala, Arcesio Bustos-Gaibor, Sandra Arias-Villon
Abstract - The problem with predicting a possible risk of fetal death is that it depends on several aspects, not only medical but also economic and social aspects of the pregnant mothers, which makes an early response to this problem very difficult. This work seeks to apply classification algorithms to detect the risk of fetal death based on socioeconomic and demographic data of pregnant mothers in Ecuador, using datasets from 2000 to 2021. Trained algorithms include decision trees, random forests, neural networks, bagging classifiers, k nearest neighbors, and naive bayes bernoulli. As cases of fetal death are very rare, over-sampling and undersampling techniques were applied to train the algorithms. The performance comparison of the trained algorithms was carried out with their respective confusion matrices. The best performance was obtained by the algorithms trained with undersampling and of all of them the performance of the neural network stood out. The best performance of the neural network was associated with its nature of classifying by assigning weights to each input parameter.