Internship engagement as a predictive pathway between an academic self-concept proxy and employability outcomes: evidence from a public undergraduate employability dataset.

Background: Employability competencies in higher education reflect both academic indicators and experiential learning. This study examines whether a behavioral internship-engagement proxy provides a predictive indirect pathway between an academic self-concept proxy and employability outcomes using the Hierarchical Attention Mediated Employability Network (HAMEN). Because the source data are cross-sectional and observational, mediation is interpreted strictly as an associational predictive decomposition rather than a causal mechanism. Methods: A secondary analysis was conducted on a publicly available Kaggle dataset containing 1,200 undergraduate records of unspecified institutional and national origin. Academic Self-Concept Proxy (ASC-P) was constructed from grade point average, core-subject score, and academic consistency, whereas Behavioral Internship Engagement Proxy (IE-P) was constructed from internship and learning-exposure indicators. Data were divided into training, validation, and test sets (70%/15%/15%); preprocessing was fitted only on the training set, and HAMEN hyperparameters were selected using five-fold cross-validation. Performance was assessed using R Results: HAMEN achieved Conclusion: HAMEN improved employability-score prediction and assigned the largest engagement attention weights to internship duration and project experience. The results indicate predictive associations only; they do not establish that academic self-beliefs or internship engagement causally produce employability outcomes because the constructs are proxy measures and the dataset provides no temporal ordering.