The global type 2 diabetes (T2D) pandemic is raising concerns about the increasing prevalence of hepatocellular carcinoma (HCC), one of the leading causes of cancer mortality in the world. In T2D patients, the causes of HCC are frequently associated with alcohol consumption. However, in addition to the patients diagnosed with alcohol use disorder (AUD), there is an undiagnosed population (subclinical alcohol use disorder, sAUD) that also contributes to the alcoholic burden of various diseases and in particular HCC. In this study, we analyzed a clinical dataset from almost 3 million T2D patients. This dataset presents several technical challenges: feature heterogeneity, high disparity, structurally missing data, and strong imbalance in the target variables. We developed methods based on XGBoost, One-Class SVM, and Multi-Layer Perceptron to identify sAUD patients who might contribute to an increase in the actual alcoholic burden on HCC. Our approaches show interesting results, allowing us to identify sAUD patients with similar clinical characteristics to AUD patients. This could suggest that the actual alcohol burden on HCC might be up to 30% higher than when considering only AUD patients. The proposed methodologies offer valuable insights not only for characterizing T2D patients but also for addressing similar challenges in other medical datasets.