Abstract: Pulmonary embolism (PE) can range from minor, asymptomatic blood clots to life-threatening emboli capable of obstructing pulmonary arteries, potentially leading to cardiac arrest and fatal outcomes. Due to this significant mortality risk, risk stratification is essential following PE diagnosis to guide appropriate therapeutic intervention. This study proposes a machine learning-based methodology for PE risk stratification, utilizing clinical data from a cohort of 139 patients. The predictive framework integrates an enhanced binary Honey Badger Algorithm (BCCHBA) with the K-Nearest Neighbor (KNN) classifier. To comprehensively evaluate the performance of the core optimization algorithm (CCHBA), a series of benchmark function tests were conducted. Furthermore, diagnostic validation tests were performed using real-world PE patient data collected from medical facilities, demonstrating the clinical significance and practical utility of the BCCHBA-KNN system. Analysis revealed the critical importance of specific indicators, including neutrophil percentage (NEUT%), systolic blood pressure (SBP), oxygen saturation (SaO₂%), white blood cell count (WBC), and syncope. The classification results demonstrated exceptional performance, with the prediction model achieving 100% sensitivity and 99.09% accuracy. This approach holds promise as a novel and accurate method for assessing PE severity.

Key words: Disease diagnosis, Feature selection, K-Nearest neighbor, Machine learning, Pulmonary embolism, Swarm intelligence')">Disease diagnosis, Feature selection, K-Nearest neighbor, Machine learning, Pulmonary embolism, Swarm intelligence