Heart failure mortality: a data driven analysis

Purpose Millions of deaths worldwide are attributable to heart failure. Unlike other chronic diseases, the patient can die within a short period of time once heart failure occurs. Hence, it is essential to provide cost-effective solutions to predict the likelihood of mortality for such patients. Rule-based classification can reveal crucial knowledge about which features are more impactful to those surviving heart failure. This study aims to investigate interpretable models to support prediction and improve survival outcomes Design/methodology/approach A potential approach to treat this issue is to use classification models developed from historical data with characterised pathological and non-pathological features. This research compares various rule-based classification algorithms using real data to identify which techniques best address the prediction of surviving heart failure. The models are evaluated for their ability to generate interpretable and clinically meaningful rules. Findings Empirical results against a real data set showed that the RIPPER algorithm produces interpretable classification models that are manageable and easy to use by medical professionals. RIPPER also demonstrates an acceptable level of predictive power. These findings indicate that rule-based approaches can balance interpretability with predictive accuracy for heart failure survival prediction. Originality/value This research highlights the role of interpretable rule-based classification in predicting survival among heart failure patients. By comparing algorithms, the study identifies RIPPER as particularly effective in producing simple and understandable models. The originality lies in demonstrating that predictive models can be both interpretable and clinically practical, providing medical professionals with transparent tools to support decision-making.