Mapping the Multidimensional Drivers of Fertility Intentions In South Korea: A Machine Learning Perspective

Background:

South Korea has experienced one of the lowest fertility rates in the world, raising serious concerns about demographic sustainability and future socioeconomic stability. In particular, fertility intentions among unmarried youths are crucial, as they directly influence future birth rates. Fertility intention should be understood as a multidimensional phenomenon shaped by multiple interacting factors rather than a single determinant.

Despite the increasing availability of large-scale data, empirical studies applying machine learning techniques to predict fertility intention among youths remain limited. Therefore, this study aims to predict fertility intention among unmarried youths in South Korea using a machine learning approach and to identify key influencing factors.

Methods:

This study uses data from the ‘2024 Survey on the Quality of Life of Youth’ conducted by the Korea Institute for Health and Social Affairs (KIHASA). The analysis focuses on unmarried youths, and fertility intention is defined as a binary dependent variable indicating whether respondents intend to have children in the future. A range of socioeconomic, demographic, political, psychological, and attitudinal variables was included.

A machine learning approach using the PyCaret framework was applied, incorporating preprocessing steps such as imputation, normalization, and multicollinearity reduction. A traditional logistic regression model was also estimated for comparison. Among several models, the Extra Trees classifier was selected based on its superior F1-score performance, reflecting a balanced trade-off between precision and recall. The superior performance of the Extra Trees classifier suggests that fertility intention is characterized by nonlinear relationships and complex interactions among variables, which are not fully captured by traditional linear models. SHAP analysis was conducted to assess variable importance and the direction of their effects on fertility intention.

Results:

Fertility intention among unmarried youths is influenced by a combination of quality of life, economic conditions, and social and attitudinal factors. The analysis was conducted using a sample of 12,823 individuals, with 8,976 used for training and 3,847 for testing. Among several models, the Extra Trees classifier demonstrated strong predictive performance, achieving an F1-score of 0.726.

Life satisfaction, social trust, employment status, housing satisfaction, and subjective health show str positive associations with fertility intention, highlighting the importance of overall well-being. Economic stability, including income and assets, also contributes positively, while residence in the metropolitan area and perceived income inequality are negatively associated with fertility intention. Overall, the findings suggest that fertility intention is not determined by economic conditions alone, but by a broader set of life experiences and social and attitudinal factors.

Implications:

These findings suggest that policies aimed at addressing low fertility in South Korea should move beyond a narrow focus on financial incentives. Improving overall quality of life, enhancing housing stability, and strengthening social trust are critical for increasing fertility intention among youths. In addition, reducing perceived inequality and fostering a more stable and supportive social environment may contribute to reversing the declining fertility trend. This is consistent with the finding that fertility intention is not determined by economic conditions alone, but by a broader set of life experiences and social and attitudinal factors.