Gender Knowledges, Cultures of Equality, and Structural Inequality: Interpreting Female Employment Patterns in Manufacturing Through Interpretable Machine Learning

Abstract

Persistent gender inequality in industrial employment continues to challenge inclusive labor systems worldwide. While education and labor market reforms have expanded opportunities for women, structural barriers remain deeply embedded in manufacturing sectors. This study adopts a systems-based perspective to investigate the institutional, demographic, and health-related factors shaping female employment in manufacturing across ten countries from 2013 to 2022. By integrating feminist political economy with interpretable machine learning techniques—including Random Forest, Gradient Boosting, and Extra Trees regressors—the study models non-linear and interactive relationships among thirteen structural indicators drawn from the World Bank’s World Development Indicators. The findings reveal that general female labor force participation is the strongest and most consistent predictor of women’s inclusion in manufacturing. Health-related variables, such as maternal mortality and fertility rates, exhibit strong negative effects, underscoring the continued influence of caregiving burdens and inadequate health systems. Education indicators show more variable impacts, suggesting that institutional context mediates their effectiveness. The use of SHAP and Partial Dependence Plots enhances the transparency of the models and supports a more nuanced understanding of how structural forces shape gendered labor outcomes. In addition to modeling structural inequalities, this study highlights how gender knowledges and cultures of equality are contextually produced and negotiated within the manufacturing sector. The findings underscore the importance of understanding both global systems and local cultural frameworks in shaping gendered employment outcomes. By linking interpretable machine learning with systems thinking, this research provides a holistic and data-driven account of industrial gender inequality. The results offer policy-relevant insights for designing more inclusive labor strategies that address not only economic incentives but also the social and institutional systems in which employment patterns are embedded.

1. Introduction

Persistent gender inequalities in industrial employment continue to raise concern across global labor systems. Despite significant improvements in women’s access to education, this progress has not resulted in equal participation in formal employment. The manufacturing sector remains one of the most segregated domains, where longstanding institutional hierarchies and social norms continue to limit women’s inclusion (Acker 1990; Charles and Bradley 2009; Baerlocher et al. 2021). In many emerging and middle-income economies, industrial growth has taken place without parallel advances in gender equality (ILO 2020; World Bank 2023). Although international agendas such as the Sustainable Development Goals (SDGs) emphasize inclusive labor markets, evidence shows that economic expansion alone does not close gender gaps in industrial work (Del Rey et al. 2021; Kabeer 2021; UNDP 2022). Understanding the reasons behind these persistent disparities remains critical for both research and policymaking.

Labor market inequality is shaped by more than differences in skills or education. It also emerges from institutional arrangements and social norms that govern access to work and determine whose contributions are recognized and valued (Fuentes and Ehrenreich 2024). Within manufacturing, workplace structures and hiring practices have historically favored men, limiting women’s opportunities despite rising levels of education (Berglind and Tarkian 2025; Fakih and Ghazalian 2015). Feminist research has highlighted the role of social and cultural capital in shaping women’s employment outcomes (Bridges et al. 2022). These forms of capital, often developed in informal or domestic settings, are rarely acknowledged in formal labor systems (Skeggs 1997; Reay 2004). Additionally, cross-national studies point to structural factors such as fertility, healthcare access, and political representation as important determinants of women’s participation in formal sectors like manufacturing (Klasen et al. 2021; Alon et al. 2022). A growing number of recent studies have begun to use machine learning to model gendered labor outcomes, particularly by incorporating demographic and institutional variables within large datasets (Jaiswal et al. 2023; Ranta and Ylinen 2023; Nuseir et al. 2021). However, many of these models emphasize prediction accuracy over interpretability and remain disconnected from broader theoretical frameworks. In practical terms, interpretable machine learning helps uncover patterns that traditional regression models often miss, such as threshold effects and non-linear dynamics. Unlike purely predictive approaches, these models not only provide accurate forecasts but also make their reasoning transparent by showing how each factor contributes to employment outcomes. This transparency allows readers who are less familiar with technical methods to see how the models connect directly to theoretical debates about gender and labor. By emphasizing these benefits at the outset, the paper highlights both the novelty and the accessibility of the approach.

This study responds to this gap by adopting a systemic perspectiveperspective that combines interpretable machine learning with feminist political economy. This research not only analyzes structural drivers but also acknowledges that knowledge about gender equality is continuously produced and experienced through the everyday practices, cultural norms, and workplace dynamics embedded in each national context. By linking structural analysis with the situated production of equality knowledges, the study addresses the complex and multi-layered nature of gendered employment patterns. Rather than treating female industrial employment as the outcome of individual or isolated factors, it is analyzed as the product of interdependent structural forces that unfold across institutional, demographic, and cultural layers. Drawing on panel data from ten countries between 2013 and 2022, the study integrates education, fertility, health, governance, and labor market indicators sourced from the World Bank’s World Development Indicators (WDI). Three ensemble models—Random Forest (RF), Gradient Boosting, and Extra Trees—are used to detect non-linear relationships and identify the most influential predictors of women’s participation in manufacturing. To enhance interpretability, Partial Dependence Plots (PDPs) are employed to visualize how key variables affect outcomes in different national contexts.

By embedding machine learning within a analytical framework, this study contributes a more holistic and theory-informed account of gender inequality in industrial employment. It bridges technical modeling with structural analysis, offering insights not only for computational social science and gender studies but also for the design of inclusive labor systems that reflect the complexity and interdependence of real-world dynamics. To guide this analysis, the study tests three interrelated hypotheses: (1) that structural indicators such as fertility, education, and governance exert significant, non-linear effects on women’s industrial employment; (2) that the influence of these factors varies across national and institutional contexts; and (3) that interpretable machine learning models can effectively uncover the systemic drivers of gendered labor outcomes. These hypotheses support a systems thinking approachthat seeks not only to model observed patterns but also to inform policy grounded in structural realities.

2. Literature Review

The underrepresentation of women in the manufacturing sector has been explored across a range of fields, including gender studies, development economics, and, more recently, computational social science. Despite significant gains in women’s education over the past few decades, manufacturing remains one of the most gender-segregated areas of employment. This gap cannot be explained by skills or qualifications alone. Cultural norms, rigid labor regulations, and institutional structures continue to shape access to industrial jobs in ways that often disadvantage women (Tejani and Milberg 2016; Kabeer 2016). Reports by the ILO (2021) and the UNDP (2022) confirm that exclusion from formal industrial work persists, particularly in middle-income and export-oriented economies, where job growth has not been accompanied by inclusive labor practices. These findings challenge the assumption that expanding education will naturally lead to more equitable employment. Instead, they point to the need for broader reforms that address the social and institutional constraints embedded in industrial labor systems.

Women’s participation in the labor market is shaped by a range of structural conditions that cannot be explained by education or personal qualifications alone. Across countries, indicators such as fertility rates, maternal health services, and access to public childcare have consistently influenced employment outcomes (Ahn and Mira 2002; OECD 2021; ILO 2020). In contexts where caregiving support is limited, women often face greater difficulties in entering and remaining in formal jobs. Kabeer and Natali (2013) emphasize that even well-designed labor policies may fall short if they do not address the institutional and social norms that define everyday gender roles. Blau and Kahn (2017) also find that wage inequality and occupational segregation remain widespread, even in high-income countries. These patterns suggest that labor market reforms alone are not enough. Sustainable progress depends on deeper transformations in workplace culture, family life, and public investment in care systems. Gender disparities in manufacturing are often maintained by rigid occupational norms and workplace cultures that continue to associate technical competence with men (Kitole 2025).

Feminist scholars have drawn on Pierre Bourdieu’s theory to explain why women with similar levels of education often face unequal outcomes in the labor market. Cultural and social capital such as personal networks, caregiving experience, and norms of respectability can shape women’s access to jobs and career progression (Huppatz 2015; Schneidhofer et al. 2021). These forms of capital usually develop in informal or domestic settings and are rarely acknowledged within formal employment systems. According to Bourdieu and Wacquant (1992), when socially meaningful traits are not recognized by institutions, individuals experience symbolic exclusion. In manufacturing, this can lead to reduced opportunities and limited support, even for qualified women (Domingo et al. 2022). These insights help clarify why structural barriers remain influential, even as educational attainment improves. They also highlight the need for studies that combine theoretical perspectives with empirical approaches in order to understand how inequality is reproduced in everyday work environments.

In addition to cultural and social capital, feminist institutionalism emphasizes that institutions themselves are gendered, shaping opportunities and constraints in systematic ways (Kenny 2007; Mackay et al. 2010). As Lorber (1994) argues, gender is not only embedded in norms and practices but is also reproduced through formal rules and organizational structures. This perspective, often referred to as “gendered institutions,” complements our focus on political economy by underlining how employment systems are structured through implicit and explicit gendered logics. Integrating this lens strengthens the theoretical basis of the study by showing how inequality is sustained not only at the cultural or individual level but also within institutional arrangements that govern labor markets.

Building on these perspectives, Connell’s concept of the gender order provides an additional theoretical lens to understand the societal structuring of inequality. The gender order describes how power, labor, and symbolic relations between men and women are institutionalized and reproduced across different settings (Connell 1987, 2002). Importantly, Connell highlights the existence of a global gender order through which men’s advantages are sustained worldwide, even though their form and intensity vary cross-nationally. To some degree, our analysis of female industrial employment across ten countries can be seen as a test of this argument: it reveals both the persistence of male advantage and the context-specific mechanisms (such as education, fertility, and institutional norms) through which gender inequality in manufacturing is maintained.

In addition to institutional arrangements, traditional gender roles remain a significant factor in shaping women’s industrial participation. In many societies, women continue to bear disproportionate responsibility for reproductive and caregiving activities, which limits their ability to take up formal industrial jobs. This pattern is particularly visible in contexts such as Mexico and Brazil, where even supportive institutional frameworks may be undermined by persistent expectations that women prioritize household and family care (ECLAC 2022; Chant 2016). These cultural norms interact with institutional barriers, reinforcing occupational segregation and restricting women’s opportunities for stable and equitable industrial employment.

Feminist historical scholarship has long emphasized the ideology of “separate spheres,” which defined men as belonging to the public and industrial domain while relegating women to the private and domestic sphere (Kerber 1988; Cott 1977). This framework not only shaped cultural expectations but also had practical consequences for women’s entry into industrial work, where their participation was often considered exceptional or temporary. These historical divisions continue to resonate in contemporary contexts through patterns of occupational segregation. Studies consistently show that women remain concentrated in a narrower set of industrial jobs, often in lower-paid or less stable positions, while men dominate technical and supervisory roles (Charles and Grusky 2004; Blau et al. 2013). Internal occupational segregation—where women and men occupy distinct roles even within the same sector—reinforces these inequalities by limiting upward mobility and recognition. Incorporating these perspectives underscores how the structural barriers we analyze are historically rooted and institutionally reproduced within industrial labor markets.

Finally, it is important to recognize that institutions do not operate in isolation. The concept of institutional interlinkages or institutional complementarity highlights how different domains such as education systems, welfare policies, and labor market regulations interact to shape gendered employment outcomes (Hall and Soskice 2001). For example, even supportive labor market policies may have limited effects if not accompanied by adequate welfare provisions or inclusive educational systems. Considering these complementarities provides a more holistic understanding of how structural inequality is reproduced within industrial employment.

In many settings, these environments reward alignment with masculine standards while failing to recognize the contributions and skills that women bring (Jenson 2024; Tayade 2022). This pattern is especially evident in export-oriented economies, where male dominance in manufacturing persists despite targeted efforts such as vocational training and flexible employment programs (Ferm and Gustavsson 2021; Seguino 2010). According to UNIDO (2021), many countries have formally adopted gender mainstreaming strategies, yet progress remains limited. Stereotypes about physical strength, long working hours, and informal hiring practices continue to shape perceptions of industrial work and often discourage or exclude female applicants (Ghosh 2021). These challenges highlight the importance of complementing policy interventions with deeper cultural change at both the organizational and sectoral level (Grogan 2023).

Studies from Turkiye and South Asia show that women’s involvement in industrial work depends not only on economic factors but also on family responsibilities, marital status, and dominant gender roles. In many of these contexts, social expectations limit women’s ability to seek steady employment or to claim equal rights in the workplace (Gündüz-Hoşgör and Smits 2008; Jayachandran 2020). As a result, women often remain less visible and have limited influence within formal industrial settings, even when they are equally qualified (Kaufman 2005). This suggests that research needs to connect cultural theory with empirical observation in order to better capture the realities of gendered labor markets (Pfau-Effinger 2012). Without a system-level understanding of how social expectations shape everyday labor dynamics, it becomes difficult to explain the persistence of structural inequalities despite policy interventions.

Many cross-country studies on female employment have used methods such as panel regression and structural equation modeling to examine the influence of education, fertility, political participation, and labor regulations (Tejani and Milberg 2016; Shittu and Abdullah 2019; Li et al. 2023; Hossain et al. 2022; Ahmad et al. 2024; Marjanović et al. 2024). While these macro-level approaches offer valuable insights, they often rely on linear assumptions and treat variables as independent, failing to account for the complex and dynamic interactions among institutional, social, and cultural factors (Najeeb et al. 2020; Gupta 2021; Klasen et al. 2021). As a result, they tend to oversimplify the processes through which gender inequality emerges and persists in labor markets. Moreover, many of these studies lack integration with social theory, which limits both the interpretability of their statistical findings and their relevance for inclusive policy development (Steiber and Haas 2012; Seguino 2010). These limitations underscore the need for system-oriented analytical frameworks that are capable of capturing non-linear relationships while remaining sensitive to broader structural and normative conditions. Approaches that combine empirical rigor with theoretical depth are essential for generating policy insights that are not only data-driven but also grounded in institutional and cultural realities. Responding to this challenge, a growing body of research has begun to apply machine learning (ML) techniques to model gendered labor outcomes more effectively (Siregar et al. 2025; Forshaw et al. 2024; Kularathne et al. 2024). These methods—such as XGBoost (version 2.1.1), Random Forest, and AdaBoost implemented via scikit-learn (version 1.4.2)—have shown improved predictive performance, particularly when working with large datasets containing demographic, economic, and institutional indicators. However, most ML applications in this field still emphasize technical accuracy over theoretical interpretability. As a result, they often fail to illuminate the social and policy-related mechanisms that shape gender inequality in employment. This gap limits their practical value for researchers and decision-makers who seek not only robust predictions but also actionable insights. Bridging this divide requires integrated approaches that connect ML methods with social theory, enabling a more nuanced understanding of how structural and normative factors jointly influence women’s employment trajectories across diverse national contexts.

In line with the increasing emphasis on system-level analysis in the literature, several recent studies have approached gendered labor inequalities not merely as isolated statistical outcomes, but as components of broader employment systems shaped by interconnected economic, institutional, and social dynamics. For example, Nica et al. (2023) examine how trade liberalization, male employment, urbanization, and foreign direct investment influence women’s participation across agricultural, industrial, and service sectors in eight South Asian economies, highlighting that trade expansion boosts female employment in industry and services but limits it in agriculture and urban contexts. In addition to this, Sobirov et al. (2024) employs dynamic ordinary least square methods to reveal how tourism expansion in Central Asia positively affects women’s employment in services and industry, while producing adverse effects in agriculture, underscoring the need for sector-sensitive, system-oriented policy frameworks. In addition, Ye and Cai (2024) analyze how the digital transformation of households fosters women’s employment in China through structural mechanisms such as enhanced human capital and reduced gender discrimination. Their system-based perspective links technological change with social outcomes, illustrating how interconnected household, labor, and information systems influence female labor force participation.

Although past research has deepened our understanding of gender inequality in manufacturing, few studies bring together cross-national data, ML methods, and feminist theory within a single analytical framework. This study aims to fill that gap by using interpretable ML to explore how structural factors shape female employment in industry across ten countries over a ten-year period. By combining quantitative modeling with theory-based interpretation, it offers a more comprehensive view of how gendered labor patterns emerge and persist. This approach contributes to both data-driven social science and research on inclusive labor policy. Building on this integrated framework, the study tests the following hypotheses:

H1. 

Structural indicators such as fertility rates, education levels, and governance quality have significant, non-linear effects on women’s industrial employment across countries.

H2. 

The impact of these structural factors varies significantly between countries with different institutional and social contexts.

H3. 

Interpretable machine learning models can effectively identify and visualize key drivers of gendered employment outcomes, offering added value beyond traditional linear approaches.

3. Methodology

This study uses a cross-national panel dataset from the World Bank’s WDI, covering the years 2013 to 2022. The sample includes ten industrializing and middle-income countries: Brazil, Germany, the Republic of Korea, Mexico, the Philippines, Thailand, Türkiye, Ukraine, the United Kingdom, and Vietnam. These countries were selected based on data availability, geographic diversity, and variation in institutional labor market structures. This targeted selection allows for a focused yet systemically diverse comparison of how structural forces shape female employment in manufacturing across different national contexts, while maintaining consistency and completeness in longitudinal data over the study period. The research design brings together feminist political economy and machine learning to model the structural factors that influence women’s employment in the manufacturing sector.

3.1. Data Collection and Variables

The dependent variable in this study is female employment in the manufacturing sector. It is measured as the percentage of women working in manufacturing relative to the total female labor force. This indicator, taken from the WDI, reflects the extent to which women are integrated into formal industrial production. Manufacturing is often seen as a marker of economic modernization, but it also tends to preserve rigid gender roles (UNIDO 2022). For this reason, the share of women in industrial employment provides a useful measure of both structural and cultural constraints. Unlike general employment indicators, participation in manufacturing often depends on a mix of education, physical mobility, social networks, and access to formal labor markets. These factors are unevenly distributed across countries and social groups (ILO 2021; Hegewisch and Gornick 2013). By modeling this outcome, the study aims to understand how national structures and social norms shape women’s entry into one of the most traditionally male-dominated sectors of the economy.

Table 1. Description of independent variables used in the machine learning model.

Code	Variable Name	Description
X1	FemaleLFParticipation15Plus	Labor force participation rate (% of women aged 15+)
X2	FemaleEmploymentTotal	Female employment rate (all sectors)
X3	FemaleLaborForceTotal	Total number of women in the labor force
X4	FemaleLFParticipation15to64	Labor force participation rate of women aged 15–64
X5	TertiaryEnrollmentFemale	Female enrollment in tertiary education (%)
X6	SecondaryEnrollmentFemale	Female enrollment in secondary education (%)
X7	EducationExpenditureGDP	Government expenditure on education as % of GDP
X8	GDPperCapita	GDP per capita (USD)
X9	FertilityRate	Total fertility rate (births per woman)
X10	FemalePopulationTotal	Total female population
X11	WomenInParliament	Female representation in national parliament (%)
X12	MaternalMortality	Maternal mortality ratio (per 100,000 live births)
X13	AntenatalCareCoverage	Percentage of pregnant women receiving prenatal care

lists the 13 independent variables used to predict female employment in the manufacturing sector. These variables reflect demographic, educational, economic, and health-related dimensions that influence women’s participation in the labor force. Each variable was selected based on its relevance in the literature and its ability to capture key structural factors affecting gendered employment across different national contexts. This study adopts an integrated modeling framework that captures the combined influence of demographic trends, educational access, institutional representation, and health infrastructure. These dimensions are treated as interconnected components of a broader system that shapes female employment in the manufacturing sector across diverse national contexts. These variables serve as the empirical basis for testing the study’s main hypotheses regarding structural determinants, cross-national variation, and the added value of interpretable machine learning approaches.

3.2. Machine Learning Framework and Analytical Strategy

To explore how structural dynamics shape female employment in the manufacturing sector, this study applies an interpretable machine learning framework grounded in feminist political economy. By modeling the share of women employed in industry across ten countries from 2013 to 2022, the analysis seeks to uncover how demographic, educational, institutional, and health-related variables interact within broader employment systems. Manufacturing is not only an economic field but also a highly structured social domain, where labor patterns reflect deeper institutional asymmetries. Thus, rather than isolating variables, we adopt a holistic view that considers employment outcomes as emergent properties of intersecting national structures.

Three ensemble learning models—RF, Gradient Boosting Regressor (GBR), and Extra Trees Regressor (ETR)—were employed to identify key predictors of women’s industrial participation. These tree-based algorithms are particularly well-suited for complex social datasets with potential non-linearities and multicollinearity, and they allow for feature-level interpretation without requiring parametric assumptions. Models were trained using z-score standardized features and optimized via five-fold cross-validation with grid search to prevent overfitting and ensure model stability across time and countries.

The data preprocessing stage involved linear interpolation to address missing values, exclusion of variables with over 20% missingness, and temporal alignment of country-level time series. Model performance was assessed through multiple metrics, including R-squared (R²), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE), to provide a multi-dimensional evaluation of prediction accuracy.

To enhance model transparency, we incorporated both PDPS and SHAP (SHapley Additive exPlanations) values. While PDPs visualize the marginal effect of each predictor across its range, SHAP values provide a more granular decomposition of how each feature contributes to individual predictions. This dual approach allows us to detect non-linear patterns, threshold effects, and interaction dynamics that are often overlooked in traditional linear models. Although the study emphasizes predictive insight rather than causal inference, these interpretable ML techniques enable a meaningful link between empirical outputs and the theoretical hypotheses articulated earlier. In doing so, we align model outputs with system-level explanations, reinforcing the study’s contribution to both policy relevance and theory-driven analysis.

By embedding machine learning within a structured analytical framework, this study not only models complex gendered employment outcomes but also contributes to ongoing efforts to integrate computational techniques with inclusive labor policy design and systemic thinking.

4. Computational Results

This section presents the empirical findings of our machine learning models, applied to cross-national panel data on female employment in the manufacturing sector. By integrating interpretable ensemble methods with structural variables across ten countries, the analysis aims to identify key predictors, assess model accuracy, and evaluate how results align with the study’s theoretical hypotheses. The findings offer both predictive insight and policy-relevant interpretations for gendered labor dynamics in industrial contexts.

4.1. Descriptives

Before presenting the results of machine learning models, we first explore the temporal and cross-national dynamics of female employment in manufacturing. These descriptive analyses provide essential context for interpreting the structural patterns observed in the modeling phase and reveal the heterogeneity that motivates a more nuanced, system-level analysis.

Figure 1 illustrates cross-national differences in changes to female manufacturing employment from 2013 to 2022. Mexico and Türkiye show substantial increases, likely reflecting policy shifts favoring labor-intensive industries and gender-inclusive labor strategies. In contrast, declines in the Philippines and South Korea may reflect economic transitions toward services or persistent gendered barriers. These variations highlight the systemic nature of labor market change and underscore the need for multi-dimensional policy approaches.

Figure 2 presents the evolving trends in female employment within the manufacturing sector over the past decade, revealing notable variation across countries. Mexico and Türkiye display consistent upward trajectories, likely reflecting labor-intensive industrial policies and strategies aimed at promoting female workforce integration. These findings align with prior studies, showing that increases in women’s industrial employment are often associated with export-oriented production models and supportive institutional conditions (Seguino 2010; Kabeer and Natali 2013).

In contrast, South Korea and the Philippines demonstrate declining trends, which may stem from automation, a transition toward service-based economies, or persistent structural barriers. These patterns are consistent with ILO (2021) findings, which emphasize that technological shifts and entrenched gender norms continue to limit women’s access to formal industrial employment in many settings. Collectively, the observed trajectories underscore the importance of national context and global industrial change in shaping gendered labor outcomes.

4.2. Model Interpretation

To evaluate the robustness and interpretability of the proposed modeling framework, this section presents a comparative performance analysis of the machine learning algorithms used in the study. Beyond assessing predictive power, we also investigate which structural factors most strongly influence female industrial employment, using model-specific insights and post hoc interpretation techniques. This dual focus enables a deeper understanding of how complex, interrelated systems shape women’s labor outcomes in the manufacturing sector across countries.

Table 2. Comparative performance of machine learning models in predicting female industrial employment.

Model	R2 Score	RMSE	MAE
Random Forest	0.93	1.085	0.742
Gradient Boosting	0.978	0.602	0.43
Extra Trees	0.98	0.585	0.436

presents the predictive performance of the ensemble-based models employed in this study. Among the algorithms tested, the ETR outperformed others, achieving the highest R² value (0.980) and the lowest RMSE and MAE scores, indicating strong predictive accuracy. These results suggest that female employment in the manufacturing sector is shaped by a combination of educational, demographic, health, and governance-related factors—interactions that are more effectively captured through non-linear modeling techniques.

To better understand the factors shaping women’s participation in industrial employment, we conducted further analysis using the ETR model. The results reveal that indicators related to general female labor market engagement, such as FemaleLFParticipation15to64, FemaleLaborForceTotal, and FemaleEmploymentTotal, play the most significant role in predicting manufacturing employment (see Figure 3). This suggests that expanding women’s access to employment opportunities overall can directly enhance their presence in the industrial workforce. In contrast, education-related and broader demographic indicators were found to be less influential in the model. Their effects may be more indirect or vary depending on national context. The predicted-versus-actual plot also supports these findings. As seen in Figure 4, most data points are closely aligned with the diagonal line, indicating that the model performs well in estimating real-world outcomes.

To improve the interpretability of the machine learning model and understand the effect of key predictors, PDPs were created for the three most influential variables identified by the ETR. Although ensemble-based models like ETR are often considered opaque, PDPs help make their predictions more transparent by showing how changes in a single variable affect the predicted outcome while keeping other variables constant.

Figure 5 illustrates that the variable FemaleLFParticipation15to64 has a strong and almost linear positive relationship with female employment in manufacturing. This indicates that as more women between the ages of 15 and 64 participate in the labor force, their presence in industrial employment increases. Similarly, FemaleLaborForceTotal and FemaleEmploymentTotal show steady upward trends, confirming that overall female engagement in the economy is closely linked to industrial workforce inclusion. These findings are consistent with previous research emphasizing the role of structural access and inclusive labor markets in shaping gendered employment outcomes (Jaiswal et al. 2023; Ranta and Ylinen 2023). Overall, the PDP analysis supports the internal consistency of the predictive model and offers valuable insights for designing policies that promote women’s participation in industrial sectors.

In addition to the PDP results, SHAP analysis was conducted to better understand how each feature contributes to the model’s predictions. As shown in Figure 6, general labor force indicators such as FemaleLaborForceTotal, FemaleEmploymentTotal, and FemaleLFParticipation15to64 have the most consistent positive impact on predicted female employment in manufacturing. Higher values of these variables are associated with stronger participation, highlighting their importance across different national contexts. On the other hand, variables like MaternalMortality and FertilityRate tend to lower the predicted values when they are high. This suggests that barriers related to health and caregiving continue to limit women’s access to industrial employment. Education-related variables, including TertiaryEnrollmentFemale and EducationExpenditureGDP, show more variable effects. This may reflect how the impact of education depends on other social or institutional factors. Overall, SHAP results support the idea that different structural indicators interact in complex ways to shape gendered labor outcomes.

5. Discussion

This study set out to explore the structural determinants of female industrial employment across ten countries using interpretable machine learning (ML) techniques embedded within a feminist political economy framework. The empirical findings lend strong support to the three hypotheses articulated earlier and reveal important patterns in how demographic, educational, and institutional indicators interact to shape gendered labor outcomes. In this section, we discuss these findings in light of the existing literature and assess their theoretical and policy implications.

Our analysis confirms that structural indicators play a crucial role in shaping women’s participation in manufacturing. Specifically, general labor market engagement variables (such as FemaleLFParticipation15to64 and FemaleLaborForceTotal) emerged as the most influential predictors in the ensemble models. These results align with previous findings that emphasize the foundational role of structural access to employment in shaping gendered outcomes (Jaiswal et al. 2023; Ranta and Ylinen 2023). The PDPs (Figure 6) further reveal strong and near-linear relationships between these indicators and female industrial employment, confirming their systemic importance.

Moreover, SHAP analysis demonstrates that health-related variables such as MaternalMortality and FertilityRate exert negative pressure on predicted female employment in manufacturing when they reach higher levels. These findings support earlier claims that caregiving burdens and inadequate maternal health infrastructure significantly constrain women’s ability to enter formal industrial work (Ahn and Mira 2002; ILO 2020). Education-related indicators, while present, had more variable and context-dependent effects, suggesting that their impact is often mediated by other institutional or cultural conditions (Kabeer and Natali 2013; Huppatz 2015). This supports the argument that improving education alone is insufficient to address labor market disparities unless complemented by changes in governance, health systems, and social norms (Tejani and Milberg 2016; Skeggs 1997).

Cross-national variation is a defining feature of our results. Countries such as Mexico and Türkiye exhibited rising trends in female industrial employment, likely due to labor-intensive export strategies and targeted gender inclusion policies (Seguino 2010; Kabeer and Natali 2013). In contrast, South Korea and the Philippines showed declining trajectories, consistent with transitions toward service-oriented economies and persistent structural barriers (ILO 2021). These heterogeneous trends affirm that institutional settings, policy environments, and cultural norms critically shape how structural indicators are translated into employment outcomes (Klasen et al. 2021; Kaufman 2005).

Furthermore, the non-linear and interactive effects observed in the SHAP summary plot suggest that the same structural variable can produce different outcomes depending on national context. For instance, tertiary education may foster inclusion in some countries but have limited or even adverse effects in others where manufacturing remains rigidly masculinized or informal (Kitole 2025; Jenson 2024). This reinforces feminist critiques that emphasize the importance of intersectional and context-sensitive analysis in understanding labor inequalities (Reay 2004; Bridges et al. 2022).

The results demonstrate that institutional and cultural dynamics shape industrial employment outcomes differently across contexts. For example, in Latin American economies such as Mexico and Brazil, persistent gender norms surrounding caregiving continue to restrict women’s opportunities even when education levels rise, whereas in OECD countries, challenges are more closely tied to occupational segregation and workplace culture. This variation indicates that structural barriers are not uniform but are mediated by national institutional arrangements and cultural expectations.

H3 hypothesis is strongly supported by the superior performance of the Extra Trees Regressor (R² = 0.980) and the interpretability gains achieved through PDPs and SHAP analysis. Unlike traditional regression models, ensemble-based ML techniques captured complex non-linearities, interactions, and threshold effects without requiring parametric assumptions. For example, the PDPs showed clear saturation points in labor force variables, indicating that marginal gains may taper off after a certain threshold—a pattern difficult to detect with linear models. SHAP values provided individualized insights, allowing us to decompose feature contributions at both global and local levels.

These results support calls in the recent literature to bridge computational social science and feminist political economy by moving beyond black-box prediction toward socially meaningful interpretation (Forshaw et al. 2024; Kularathne et al. 2024). The combination of technical robustness and theoretical relevance in this study demonstrates that interpretable ML methods can serve as powerful tools for understanding and addressing structural gender inequality, particularly when grounded in systemic thinking.

Broader Implications and Contributions

The findings contribute to an emerging consensus that gender disparities in manufacturing are sustained by interacting structural forces, not merely by skill or education gaps (Baerlocher et al. 2021; Charles and Bradley 2009). Our ML approach validates the theoretical claim that female industrial employment is an emergent property of institutional arrangements, demographic pressures, and social norms operating across multiple levels (Fuentes and Ehrenreich 2024; Nica et al. 2023). In doing so, the study aligns with prior sectoral analyses that emphasize the importance of integrating labor policy with health, education, and governance reforms (Ye and Cai 2024; Sobirov et al. 2024).

These findings also suggest that the pursuit of gender equality in manufacturing cannot be separated from the cultural and historical processes through which equality norms and gender knowledge are constructed, challenged, and transformed. A feminist and decolonial perspective, as adopted in this study, reveals how institutional reforms must be accompanied by shifts in everyday practices, symbolic recognition, and local understandings of equality.

Moreover, the limited role of educational indicators in the models reinforces critiques of human capital theory that assume a linear translation from education to employment (Blau and Kahn 2017; Domingo et al. 2022). Instead, our findings underscore that even highly educated women may remain excluded from manufacturing jobs unless broader systems of care, representation, and institutional recognition are addressed. This echoes the work of Bourdieu and feminist theorists who argue that symbolic exclusion and cultural devaluation are central mechanisms of labor market inequality (Skeggs 1997; Schneidhofer et al. 2021).

6. Conclusions

This study examined the structural factors that shape female employment in the manufacturing sector across ten countries between 2013 and 2022. By combining interpretable machine learning models with a feminist political economy perspective, the research identified key drivers of gendered labor outcomes and offered a broader understanding of how institutional and demographic factors influence women’s participation in industry.

The findings show that women’s overall participation in the labor market, especially among working-age groups, is closely linked to their inclusion in manufacturing. However, this relationship is shaped by wider systems, including healthcare access, caregiving responsibilities, and political representation. High fertility rates and maternal mortality remain significant barriers, pointing to the importance of investing in care infrastructure and health services. Education continues to matter, but its impact varies depending on context and is not enough on its own to improve industrial employment outcomes for women.

This study also demonstrates that interpretable machine learning can uncover complex patterns that traditional methods often miss. The use of SHAP and PDP tools helped reveal how different variables interact and influence outcomes in non-linear ways. These techniques allowed for a deeper understanding of how structural conditions affect labor markets and offered insights that are useful for designing better policies.

The results suggest that there is no single solution for promoting gender equality in industrial work. Policies need to be adapted to each country’s specific context and supported by changes in care systems, workplace practices, and public institutions. Efforts that focus only on education or economic incentives are unlikely to succeed if they do not address the broader systems in which women live and work.

From a policy perspective, these findings suggest that one-size-fits-all strategies are unlikely to succeed. In countries where traditional gender roles remain influential, expanding public childcare and promoting shared parental leave may be more effective than general labor reforms. By contrast, in higher-income contexts where women are already integrated into industrial employment, stronger enforcement of anti-discrimination laws and measures to reduce internal occupational segregation may yield greater progress. Future research should continue to investigate how national differences condition the effectiveness of policy interventions.

More importantly, this study shows in simple terms why interpretable machine learning matters for social science. It allows us to move beyond “black box” models and see how education, health, and institutional variables interact in real-world settings. The novelty lies not only in the use of advanced algorithms, but in the way these tools are translated into understandable insights that can inform both theory and policy. Presenting results in this way ensures that the technical procedures remain accessible to a broader audience, including scholars and practitioners who may not specialize in computational methods.

Overall, the study demonstrates that effective policy and practice must engage with both the structural systems and the cultures of equality that shape women’s industrial employment. Future research should further explore how gender knowledge is produced, negotiated, and enacted across diverse cultural and institutional landscapes.

In conclusion, this research connects data-driven analysis with social theory to provide a more complete picture of gender inequality in industrial employment. It shows that structural conditions matter, and that thoughtful use of machine learning tools can support both academic research and policy development. Future studies could further probe how institutional interlinkages shape women’s industrial employment. In particular, examining how education systems, welfare regimes, and labor market institutions complement or contradict one another may reveal additional pathways through which gender inequality persists. Such an agenda would extend the present study by situating structural determinants within broader institutional ecosystems.