Objective: 22q11.2 deletion syndrome (22q11DS) is the leading genetic cause of palatal anomalies, presenting a heterogeneous spectrum of clinical manifestations that complicates early diagnosis. No study has systematically investigated the co-occurrence of symptoms spanning cardiovascular, neuropsychiatric, craniofacial, immune, endocrine, renal, and familial domains within a single cohort using unsupervised learning methods for 22q11DS stratification. This study proposes a pipeline to identify subgroups and associated symptom patterns from binary clinical data, aiming to improve screening and referral for genetic testing. Design: Cross-sectional, retrospective study comparing three unsupervised clustering frameworks, K-medoids, DBSCAN, and Latent Class Analysis (LCA), followed by association rule mining applied to individuals with confirmed deletion. Setting: Brazilian Database of Craniofacial Anomalies (BBAC), a national clinical registry. Participants: A cohort of 297 individuals with suspected 22q11DS. Main Outcome Measures: Cluster separability (silhouette coefficient), characterization of phenotypic subgroups, and sensitivity/specificity of the current screening protocol stratified by subgroup. Results: LCA achieved the highest cluster separability (silhouette = 0.165), yielding four distinct subgroups: Cardiac-predominant (C0), Craniofacial-palatal (C1), Neurodevelopmental-psychiatric (C2), and Mild- craniofacial (C3). Cluster coherence was confirmed by medical specialists. Quantitative protocol assessment revealed adequate overall sensitivity, with markedly lower performance for C2 and C3, identifying profiles at higher risk of being missed by current screening criteria. Conclusions: These findings provide a data-driven basis for recalibrating symptom weights and referral thresholds, improving sensitivity for underperforming subgroups while reducing unnecessary referrals, optimizing the balance between diagnostic coverage and the financial burden of genetic testing.