Abstract
Heart failure is a complex medical condition influenced by various clinical factors. The diversity of patient data requires a more advanced, technology-based approach to support faster and more accurate diagnoses. This study investigates the effectiveness of combining Random Tree and Forward Feature Selection (FFS) to improve predictive accuracy in diagnosing heart failure an approach that remains relatively underexplored in existing heart failure research. This study evaluates the effectiveness of combining FFS and the Random Tree algorithm in improving the accuracy of heart failure classification using clinical data from the UCI repository. Performance is assessed using standard classification metrics through 10-fold cross-validation. Results show that FFS significantly enhances model performance, increasing accuracy from 75.56% to 84.28% and AUC from 0.632 to 0.786. This combination proves effective for decision support systems in medical diagnostics. Although recall experienced a slight decline, the overall improvement in classification metrics such as accuracy and precision, demonstrates that FFS effectively enhances model focus. These findings indicate that the integration of FFS and Random Tree yields a robust classification framework, offering practical potential for clinical decision support systems in heart failure diagnosis.