1. Introduction

The infant mortality has long been recognized as a significant marker of a society’s overall health and well-being reflecting both medical care quality and broader social conditions [1]. One of the principal objectives of the United Nations' Sustainable Development Goals is to decrease infant mortality. Goal 3, in particular, which aims to reduce the global neonatal mortality- defined as infants die within the first 28 days after birth-to below 12 per 1,000 live births by 2030 [2]. United States remains the far higher infant mortality rate and other pregnancy related death rates than that of other high-income nations, which indicates possible persistent health disparities in the US [3]. Despite progress made in many parts of the world, according to CDC, though there have been a few minor increases, 2022 marked the first statistically significant rise since 2002 [4]. Different possible socioeconomic factors might also contribute to disparities in infant death rates, in addition to particular medical issues that have an important effect (such as congenital malformations, a medical acronym for birth defects). For instance, in contrast to non-Hispanic White women, African American women experience a three to four times greater danger of dying during childbirth or from complications associated with childbirth [5]. There is also study shows that there was a rising trend in overall term infant mortality as rurality increased, primarily due to higher post-neonatal mortality rates in more rural areas [6].

According to research published in BMC Public Health, more than 60% of deaths in the first week is preventable [7]. To better quantify and diminish risks, many scholars have employed a variety of methods to analyze the factors influencing infant mortality rates. Nasejje used random forest approach tracking socioeconomic factors of under-5 mortality rate(U5MR) in sub-Saharan Africa with a mean C-index value greater than 0.5, suggesting well performing in determining factors of U5MR [8]. Similarly, Jennifer Lee applied random forest model to predict mortality risk for preterm infants with high accuracy of 88% [9]. Okelue utilized CDC Wonder database to analyze the trends and factors associated with infant death [10]. CDC Wonder, created by the Centers for Disease Control and Prevention (CDC), is a comprehensive information and communication platform designed for public health purpose [11]. Inspired by these scholars, in this study, the author will utilize data from the CDC Wonder database, along with a Random Forest model, to identify and quantify the socioeconomic factors influencing infant mortality rates. The importance of these factors will be ranked based on their impact.

2. Methodology

2.1. Data source

The CDC Wide-Ranging Web Data for Epidemiologic Research (CDC WONDER) database, an extensive public health information platform, served as the study's data source. Data encompassed from 2017-2022 years, offering a robust and recent sample for analyzing trends in infant mortality.

2.2. Variable selection

This study aims to rank the importance of various socioeconomic factors, and therefore multiple independent variables have been selected, including the mother's education level, race, and age; the father's education level, race, and age; urbanization status; birthplace; Census Region; Source of Payment for Delivery; and Medical Attendant, among others. Dependent Variable is Infant Mortality Rate. According to CDC, the term “infant mortality” refers to an infant passing away before turning one year old (0-364 days).

2.3. Model selection

In this study, a Random Forest model was chosen for its effectiveness in handling high-dimensional datasets, especially ranking the importance of different socioeconomic factors. The Random Forest ensembles machine learning technique is employed to generate multiple decision trees, and the forecasts are subsequently combined in order to minimize over-fitting and increase precision. Its ability to handle categorical variables allows it to effectively rank the factors based on their contribution to infant mortality.

However, due to the model's limitations, only five factors can be compared at a time. Therefore, the Random Forest model was applied multiple times in this study to analyze various factors. Cross-comparisons were conducted on the higher-weighted factors to further determine their significance.

3. Results and discussion

3.1. Model results

There are three rounds of use of random forest model. Here are the results in Figure 1. In the first round of Random Forest analysis, five economic factors were selected: mother's education level, father's education level, mother's single race, father's single race, and mother's age. The feature importance plot (Figure 1) reveals that mother's education emerged as the most significant factor, with a feature importance score of 0.4, indicating a strong correlation between the mother's education background and infant mortality. Father’s single race ranks the second place with moderate score of 0.25, indicating an important correlation as well.

Figure 1. First round: Feature Importance of Five Factors.

As shown in Figure 2 and Table 1, mother’s education level is defined as the most important factor in the first round, this variable was further analyzed to understand its relationship with infant mortality rates. The trend in Figure 2 shows a clear inverse relationship between the mother’s level of education and the infant mortality rate. This result aligns with existing study that higher maternal education is often associated with better access to healthcare, resulting in better child health outcomes.

Figure 2. Infant death rate associated with mothers’ education levels

Table 1. Variable explanation

The result of second round is as below:

Figure 3. Round 2: Feature Importance of Five Factors

The second round of Random forest analysis, as shown in Figure 3, highlights a significant shift in the influence on infant mortality. Father’s age stands out as the most critical factor, with a feature importance score of 0.68. This finding shows that Infant mortality is significantly influenced by the father's age, with this effect substantially outweighing that of other factors.

In comparison, mother's education level and mother's age both show moderate importance, with scores of 0.11 and 0.10, respectively. While maternal factors are still relevant, the relatively lower importance scores indicate that paternal age has a stronger impact in this model run.

Figure 4. Infant Death Rate Associated with Father’s Age

As further analyze the influence of father’s age, The plot (figure 4) presents a U-shaped relationship between paternal age and infant mortality rates, with rates being highest for fathers in the youngest age group (15-19 years old) and older age groups (above 50 years old). This may suggest that fathers in 30-39 years old tend to be in more stable social, financial, and health situations. The result of third round is shown below:

In the third round of the analysis, as displayed in Figure 5, the birthplace of the infant is the most influential factor, with a feature importance score of 0.51. According to the CDC WONDER database, the infant death rate per 1,000 live births is 5.61 in hospitals, 1.67 in freestanding birth centers, and 6.68 in home births (residences).

Figure 5. Feature Importance of Four Factors

3.2. Model evaluation

To guarantee the validity of the findings, three independent evaluations of the random forest model were conducted. The evaluation outcomes are listed below.

Table 2. Model Evaluation for Round 1

In the first round as shown in Table 2, the model shows strong performance overall, with high R² and EVS values and relatively low error metrics (MAE, MSE, RMSE). Though the higher MAE and MAD on the test set suggest that there might be some instances in which the model generalizes effectively to new data, even in cases when the predictions of the model differ from the actual values.

Table 3. Model Evaluation for Round 2

As can be seen from the Table 3, in the second round, despite the drop in performance, the test set R² value of 0.787 and the explained variance score of 0.787 indicate that the model is still reasonably good at predicting infant mortality on new data. However, the higher test errors (MAE, MSE, RMSE) suggest that there are some large prediction errors in the test set, which suggests that father’s age may be not that predominant.

Table 4. Model Evaluation for Round 3

The model evaluation for round 3 is in Table 4. The model generally performs well, with minimal signs of over-fitting. The close performance on training and test sets, as indicated by R², MSE, and RMSE, suggests the model is reliable.

4. Conclusion

Based on the information discussed in this article, it can be said that the age of the father, the mother's educational attainment, the infant's place of birth, and the father's race are the socioeconomic characteristics that have the biggest effects on infant mortality in the US. For the father's age, its impact on infant mortality shows a U-shaped curve as the father's age increases. Regarding the mother, there is a general negative correlation, meaning that the infant mortality rate goes down with increasing maternal education. As for the place of birth, the mortality rate is significantly lower for infants born in freestanding birth centers, while the mortality rate for home births is slightly higher than that of hospital births.

At the national policy level, the government should focus on above socioeconomic factors influencing infant mortality. First, improving maternal education through greater investment in education, particularly for disadvantaged groups, can significantly reduce infant mortality. For instance, establishing scholarships for women in impoverished areas, offering educational seminars for women who have not completed formal education, and providing free health check-ups for pregnant women. Additionally, expanding the accessibility of high-quality prenatal and postnatal service, especially in rural and underserved areas, is crucial. Given the U-shaped relationship between paternal age and infant mortality, targeted health interventions and education for younger and older fathers are needed. Finally, initiatives to decrease racial disparities in healthcare access and outcomes should also be be given top priority in order to ensure equal opportunities for all families.