Asthma Risk Classification in Pediatrics Using Gaussian Naive Bayes

Abstract

Pediatric asthma remains a significant public health concern, often complicated by challenges in early detection and consistent monitoring. Assessing asthma risk in pediatric patients, particularly those under five years of age, is challenging because traditional diagnostic tests often require patient cooperation and precise breathing maneuvers, which are difficult for young children to perform. Caregivers may also have varying levels of knowledge and different perceptions regarding the severity of asthma risk. To enable rapid and non-invasive triage, this study proposes an asthma risk classification approach using a machine learning model based on the Gaussian Naive Bayes (GNB) algorithm. The model was trained on synthetic data designed to approximate real-world physiological patterns, with heart rate and peripheral blood oxygen saturation (SpO2) serving as the key input features and asthma risk levels (high, moderate, low) as the target classes. The trained GNB model achieved an overall classification accuracy of 86% on the test set. Evaluation using the F1-score revealed 0.94 for the low-risk class (Precision: 0.92, Recall: 0.96), and 0.78 for the high-risk (Precision: 0.88, Recall: 0.70), while the moderate-risk class achieved an F1-score  of 0.77 (Precision: 0.75, Recall: 0.80). GNB also  the Decision Tree (81.00%) and Random Forest (81.00%) baselines and slightly exceeded the performance of the Support Vector Machine (85.00%). Furthermore, the evaluation of model behavior showed that accuracy gap between the training and testing results remained below 0.03 toward the end of the learning curve, indicating limited overfitting on the synthetic dataset.