Bridging Educational Inequity in Nepal through Explainable AI and Social Theory Integration

Abstract

This research seeks to address persistent socioeconomic disparities in Nepal’s education system by integrating explainable artificial intelligence (XAI) with foundational social theories. While enrollment rates have improved, inequities in access, retention, and learning outcomes remain among communities marginalized by caste, gender, and geography. Existing research and policies often depend on outdated statistical approaches and fail to combine social theory with modern machine learning. To overcome this gap, we adopt a mixed-methods design that blends quantitative modeling with qualitative insights from educators, policymakers, and community stakeholders. Using national datasets (EMIS, NLSS), machine learning models such as Random Forest and XGBoost are applied to predict educational disparities. SHAP (SHapley Additive Explanations) is employed to interpret results and highlight the most influential factors. These patterns are further contextualized using Sen’s Capability Approach and Bourdieu’s Cultural Capital Theory, ensuring that findings reflect both structural conditions and lived experiences. The study delivers several policy-relevant outcomes: a resource allocation framework to support equitable distribution, interactive dashboards for simulating policy scenarios, and earlywarning indicators for student dropouts. Importantly, the qualitative component complements the quantitative models, capturing voices and perspectives often excluded from policy discussions. By linking XAI with equity-focused theories, this work contributes to academic debates on educational data science, while also providing actionable tools for policymakers. Ultimately, it supports evidence-based advocacy that empowers marginalized communities and advances a more inclusive education system in Nepal.