Analysis of Early Age Fertility in Women using Machine Learning Survival Estimation Approach

Abstract

Early age fertility is a significant public health concern in Pakistan, with profound implications for women’s health, socio-economic development, and population dynam ics. This study explores the socio-demographic factors influencing early age fertility among women in Pakistan using the Pakistan Demographic and Health Survey (PDHS) dataset from 2017-2018. The research aims to identify the key socio-demographic de terminants of early age fertility and to evaluate the effectiveness of advanced survival analysis models in predicting these outcomes. The study addresses two primary objectives: first, to identify and analyze socio demographic factors associated with an increased risk of early age fertility, and second, to compare the predictive performance of three advanced survival models—the Cox Proportional Hazards Model (CPH), Random Survival Forest (RSF) Model, and Con ditional Inference Forest (CIF) Model—in the context of early age fertility prediction. To achieve the first objective, the study employs the CPH model to assess the impact of various socio-demographic factors on early age fertility. The results indicate that lower educational attainment, rural residence, and lower socio-economic status are significantly associated with an increased risk of early age fertility. Specifically, women with no education have a 4.858 times increased risk of early age fertility compared to those with higher education, and those living in rural areas face a 1.123 times greater risk compared to their urban counterparts. Additionally, women from poor socio economic backgrounds are 1.229 times more likely to experience early age fertility than those from rich backgrounds. For the second objective, the study compares the performance of the CPH, RSF, and CIF models in predicting early age fertility outcomes based on two variables: Age at Marriage (AAM) and Age at First Birth (RAAFB). The CPH model proves to be the most effective for predicting early age fertility related to AAM, as it exhibits the lowest prediction error and Integrated Brier Score (IBS), along with a high C-index indicating reliable predictions. Conversely, for the RAAFB variable, the CIF model is identified as the best model due to its lowest prediction error rate and IBS score, despite RSF’s higher C-index. The CIF model’s superior performance in terms of error rate and IBS demonstrates its capability for precise prediction of early age fertility outcomes, making it the preferred choice for this aspect of the analysis. This study contributes to the understanding of early age fertility in Pakistan by identifying critical socio-demographic factors and demonstrating the effectiveness of advanced statistical models for predicting early age fertility. The findings emphasize the need for targeted interventions addressing educational disparities, rural-urban dif ferences, and socio-economic inequalities to mitigate the risks associated with early age fertility. Additionally, the study highlights the importance of using comprehensive performance metrics for model selection, with CIF emerging as the optimal model for precise prediction of early age fertility outcomes based on RAAFB.