PREDICTIVE ANALYTICS FOR STUDENT PERFORMANCE IN DIGITAL MEDIA COURSES

The fast adoption of the digital media technologies has altered the practices of learning, creative production, and assessment in higher education. Traditional evaluation methods are mostly retrospective and do not offer much assistance in the early detection of academic vulnerability. This paper seeks to solve this problem of poor predictive visibility by suggesting a predictive analytics model of student outcomes in digital media learning. The methodology uses feature engineering and normalization, time-series trend and supervised machine learning classification, feature engineering and normalization, model explainability. Models to be used are Random Forest, Gradient Boosting, and Logistic Regression to analyze multisource data, such as assignment submissions, logs on digital tool use, attendance, assessment scores, peer collaboration measures, and creative project assessment, and SHAP-based interpretability to analyze feature impacts. According to experimental findings, the given framework is able to reach an average prediction accuracy of 92.56 and the values of precision and recall of 91.48% and 92.02, respectively. Early-warning predictions of mid-semester are 86.94% accurate, which can be used to implement academic interventions in time. The analysis of contribution to a feature indicates that the contribution of project iteration frequency, degree of engagement, and punctual submission is more than 64% predictive. The results prove that predictive analytics would be a useful tool to facilitate early intervention, personalized comments, and data-driven instruction methods. The results of the research are that the AI-based performance prediction improves the performance of learners and the planning of instruction.