Outcome-Based Education (OBE) requires that specific Course Outcomes (COs) are clearly linked to broader Program Outcomes (POs) to meet accreditation standards and support ongoing curriculum improvement. However, manual CO–PO mapping is time-consuming, inconsistent, and prone to bias. This study evaluates Support Vector Machines (SVM) and Random Forest (RF) for automated CO–PO alignment using real mappings from Ghazi University, validated by faculty members. We enhance traditional TF–IDF features with binary indicators derived from Bloom’s Taxonomy and Biggs’s constructive alignment, while addressing class imbalance through SMOTE. Models are assessed using ten-fold cross-validation and a held-out test set, with performance measured by macro-averaged precision, recall, F₁-score, and discipline-level fairness metrics. SVM achieves an F₁-score of 0.89, outperforming RF’s 0.85 (McNemar’s p = 0.00014). An active-learning loop routes low-confidence predictions to faculty, maintaining Cohen’s κ = 0.72 alignment with human judgment. These findings demonstrate the potential of automated approaches to transform CO–PO mapping in OBE systems.