Explainable AI and Machine Learning for Exam-based Student Evaluation: Causal and Predictive Analysis of Socio-academic and Economic Factors

Multiple factors beyond effort influence student success: The study identified that academic performance depends on a complex combination of socio-academic and economic factors—including study hours, scholarships, parental education, and prior academic performance—rather than any single cause, suggesting interventions should address this multifaceted reality.

AI models can reliably predict student performance: Ridge Regression achieved a Mean Absolute Error of only 0.12 when predicting student CGPA, demonstrating that machine learning can provide accurate forecasts of academic outcomes that could help identify at-risk students early.

Explainability is essential for fair educational AI: The researchers used explainable AI techniques (SHAP, LIME, and Interpret) to ensure their models' decisions were transparent and interpretable, which is critical for responsible deployment in education where decisions affect students' futures and should be understandable to educators and students alike.

Economic and family background significantly shape educational outcomes: The analysis highlighted that scholarships and parental education are among the critical factors influencing CGPA, underscoring how socioeconomic disparities can affect academic performance and suggesting the need for targeted support for disadvantaged students.

Personalized insights can empower student decision-making: The resulting web-based application provides students with individualized predictions and identifies specific areas for improvement, representing a responsible AI application that gives students agency to act on algorithmic insights rather than simply being evaluated by them.