Multi-Label Learning Model for Diabetes Disease Comorbidity

Multi-label modeling of clinical data is a challenging classification problem, especially for diseases with comorbidities. The complexity of the dataset makes it difficult to detect hidden patterns and infer useful information about disease classes that can occur simultaneously or successively. The presence of comorbidities has a significant impact on the treatment and management of diseases like diabetes. Hence, this research aims to build an intelligent clinical prediction model tool with a multi-label classification (MLC) algorithm for medical professionals to find trends in the patient’s data that show threats related to specific chronic illnesses. Patient’s clinical information consists of 150,137 anonymized records with 214 but regulated to 147 attributes and 8 labels: ‘lymphoma,’ ‘aids,’ ‘leukemia,’ ‘cirrhosis,’ ‘immunosuppression,’ ‘diabetes mellitus,’ ‘solid tumor with metastases’ and ‘hepatic failure.’ The dataset was split into a 70:30 train-test ratio. Tenfold cross-validation was used to assess the projection accuracy with ranking, example, and label-based metrics. This research proposed BaggingML, PS, PSt, DeePML, and RAkEL MLC models with a decision tree (DT) as a base classifier to learn the study dataset and compare their results based on standard metrics.