Automated detection of determining factors of pregnancy in adolescence

Abstract

Introduction: pregnant adolescents are less likely to form stable romantic relationships and are more likely to suffer emotional disorders. They are also more susceptible to various complications during pregnancy and childbirth.

Objective: to evaluate machine learning techniques to determine risk factors for teenage pregnancy.

Methods: a research with a causal correlational design was carried out. The data was obtained from the Demographic and Family Health Survey - ENDES 2021 National and Departmental, which covered the years 2018 to 2020. At the time of the interviews, their database contained information on 16,825 Peruvian adolescent women aged 12 to 19 years, who constituted the study universe. Nine algorithms were implemented: support vector machine, binary logistic regression, decision tree, adaptive boosting (AdaBoost), gradient boosting, extreme gradient boosting (XGBoost), extremely random trees (ExtraTrees), bootstrap aggregation, and random forest. Their metrics were considered as variables to be taken into account in the evaluation, precision, and the area under the curve.

Results: The most accurate algorithm was the random forest (0.965825), followed by gradient boosting (0.963744), decision tree, and support vector machines (0.963155, both).

Conclusions: the random forest was the most accurate technique; in addition to the identification of the factors in question, the three most important ones were distinguished. This study is a valuable precedent for the application of machine learning techniques in the prediction of various variables necessary to improve public management