Machine Learning for Student Success Prediction in Maritime Education at STIP Jakarta

Abstract

Maritime training institutions experience average student attrition rates of 18-25% annually, representing substantial waste of educational resources, lost tuition revenue, and unmet workforce development objectives, yet most institutions rely on reactive interventions occurring after academic failures rather than proactive support preventing dropout through early risk identification. This research presents the design and validation of machine learning models predicting student dropout probability and academic struggle enabling targeted interventions at Sekolah Tinggi Ilmu Pelayaran Jakarta. Employing design science research methodology with qualitative stakeholder evaluation, the study engaged academic advisors (n=10), former dropout students (n=12), and intervention specialists (n=8) through structured interviews examining prediction accuracy, intervention effectiveness, and ethical considerations. Random forest ensemble models processing 127 predictor variables from pre-enrollment characteristics, academic performance trajectories, engagement metrics, and psychosocial indicators achieved 87% dropout prediction accuracy with 84% recall identifying at-risk students average 8.3 weeks before potential withdrawal. Thematic analysis revealed strong support for predictive analytics coupled with privacy concerns, identifying critical themes of early intervention enablement, resource allocation optimization, and student welfare protection. Pilot implementation with 620 students demonstrated 42% dropout reduction (from 21% to 12%), $1.4 million annual tuition revenue recovery, and 56 additional graduates annually, contributing validated ML architectures and empirical evidence supporting intelligent student success systems in maritime vocational training contexts addressing retention challenges and workforce pipeline optimization.