Application of Data Science to Profile and Predict Divorce in Ecuador

Abstract

This research combines data science techniques, such as cluster analysis, and predictive modeling to deeply understand divorce profiles in Ecuador and predict their occurrence within the first 15 years of marriage. Historical divorce data published by the Ecuadorian Institute of Statistics and Census (INEC) were used, which include sociodemographic variables such as age at marriage, number of children, educational level, nationality, and years of marriage.

Supervised and unsupervised models were applied. The former allowed for the identification of common patterns among divorced couples, which enabled a better understanding of the variables influencing separation decisions. The DBSCAN model revealed profiles associated with short-term marriages, young couples, and relationships with few children. The supervised XGBoost model was then trained to predict divorce after ≤ 15 years of marriage, which predicted that the highest risk group for divorce is spouses between the ages of 25 and 40. This research contributes significantly to the understanding of divorce in Ecuador, providing valuable information for decision-making in various fields, such as public policy design, social intervention, and prevention. It can also support personalized counseling in institutions and therapists, and provide effective guidance to at-risk couples. Finally, this article will lay the groundwork for future studies in this area, promoting the use of data science to address relevant social problems.