Curriculum innovation remains a critical challenge in computer engineering education due to rapidly evolving technological demands and industry requirements. Traditional curriculum evaluation approaches, which rely on periodic reviews and manual assessments, are often insufficiently adaptive and lack predictive capabilities. This systematic review examined the role of machine learning (ML) and predictive analytics in curriculum evaluation and innovation, with the goal of developing an evidence-based framework for continuous, data-driven curriculum improvement. Following PRISMA 2020 guidelines, a structured search was conducted across five databases (IEEE Xplore, Scopus, Web of Science, SpringerLink, Google Scholar) for peer-reviewed studies published between 2020 and 2025. After screening 342 initial records, 41 studies met the inclusion criteria. The Mixed Methods Appraisal Tool (MMAT) was used for quality assessment. Findings revealed that while ML techniques particularly classification models (Random Forest, Support Vector Machine), deep learning architectures, and ensemble methods are widely used for predicting student performance and retention (82.9% of studies), there is limited integration of these insights into curriculum redesign processes (only 17.1% address curriculum-level action). A significant operational gap exists between prediction and actionable curriculum improvement. Based on the synthesis, this study proposes an ML-driven framework that integrates data preprocessing, predictive modeling with explainable AI (XAI), insight generation, and curriculum decision support. Key best practices include interpretable-by-design approaches, stakeholder integration, and addressing ethical fairness.

Keywords: curriculum innovation, computer engineering education, Machine Learning (ML), predictive analytics, systematic review, PRISMA 2020, Mixed Methods Appraisal Tool (MMAT)