Background & Objective: Disproportionality analysis is conventionally used for adverse event (AE) signal detection in spontaneous reporting data; however, its limitations in handling incomplete and class-imbalanced data remain. Knowledge graph (KG)-integrated machine learning offers a potential solution by capturing multidimensional drug-AE relationships. This study aimed to evaluate its applicability using CDK4/6 inhibitors as a case example. Methods: Using FAERS data (2024 Q1–2025 Q3), we constructed a KG centered on CDK4/6 inhibitor–related AEs (palbociclib, ribociclib, abemaciclib), enriched with mechanism-of-action and target data from ChEMBL and UniProt (10,379 nodes; 316,337 edges). Network-based features—including AE specificity, KG label adjacency, AE PageRank—were used to train a Random Forest classifier. Results: Despite class imbalance (positive labels: 13.2%), the model achieved an F1-score of 0.62. High-priority signals for palbociclib—including heart failure and pulmonary embolism—were undetected by conventional disproportionality analysis, consistent with CDK4/6 inhibitor-associated venous thromboembolism. Ribociclib signals such as acute kidney injury were also identified exclusively by the model. Conclusion: This KG-integrated framework complements pharmacovigilance by capturing multidimensional drug-AE relationships, demonstrating robustness under class imbalance, and offering scalability for oncology safety surveillance.