AN INTEGRATED CLUSTERING AND DEEP LEARNING FRAMEWORK FOR EDUCATIONAL PERFORMANCE ASSESSMENT IN PRIMARY AND SECONDARY EDUCATION

Abstract

Primary and secondary education systems generate large-scale survey and monitoring data from students, teachers, school leadership, and external observers; however, most analyses remain descriptive and do not provide actionable performance assessment or targeted intervention planning. This study proposes an integrated clustering and deep learning framework to address this gap by discovering stakeholder profiles and predicting educational quality outcomes under a unified pipeline. The framework first performs data cleaning through numeric conversion, removal of empty variables, median imputation, and z-score standardization. Unsupervised clustering is then applied to segment stakeholders into meaningful groups using K-Means, DBSCAN, and Agglomerative clustering, enabling differentiated policy actions and identification of outlier profiles. Subsequently, a deep learning model (ANN/MLP) is trained to learn non-linear relationships among educational indicators for performance assessment and risk classification. Experiments are conducted on the MP Education Survey dataset, comprising five aligned datasets primary students, secondary students, teachers, headmasters, and observers linked by school-level identifiers. Clustering results reveal distinct ability, governance, and infrastructure condition groups, while DBSCAN highlights anomalous schools and teacher profiles. Deep learning achieves consistently strong classification performance across stakeholder datasets (approximately 0.81–0.88 accuracy), with the best performance observed for leadership and teacher quality prediction. The proposed framework supports profile-based interventions and scalable monitoring for educational improvement.