Abstract
Biomedical datasets constitute a rich source of information, containing multivariate data collected during medical practice. In spite of inherent challenges, such as missing or imbalanced data, these types of datasets are increasingly utilized as a basis for the construction of predictive machine-learning models. The prediction of disease outcomes and complications could inform the process of decision-making in the hospital setting and ensure the best possible patient management according to the patient's features. Multi-label classification algorithms, which are trained to assign a set of labels to input samples, can efficiently tackle outcome prediction tasks. Myocardial infarction (MI) represents a widespread health risk, accounting for a significant portion of heart disease-related mortality. Moreover, the danger of potential complications occurring in patients with MI during their period of hospitalization underlines the need for systems to efficiently assess the risks of patients with MI. In order to demonstrate the critical role of applying machine-learning methods in medical challenges, in the present study, a set of multi-label classifiers was evaluated on a public dataset of MI-related complications to predict the outcomes of hospitalized patients with MI, based on a set of input patient features. Such methods can be scaled through the use of larger datasets of patient records, along with fine-tuning for specific patient sub-groups or patient populations in specific regions, to increase the performance of these approaches. Overall, a prediction system based on classifiers trained on patient records may assist healthcare professionals in providing personalized care and efficient monitoring of high-risk patient subgroups.