Leveraging Machine Learning for Early Detection of Asthmatic Children in Healthcare

Abstract

Breathing problems are common and often transient in early childhood, making it challenging to predict which children will develop persistent asthma. Early and accurate diagnosis is important to ensure appropriate medical treatment. Current prediction models, based on small and specific sample groups, demonstrate limited precision. Machine learning (ML) techniques, however, show promise for providing more accurate and generalizable predictions compared to traditional models. Method: In this study, we developed ML-based prediction models for childhood asthma using a health dataset. Dimensionality was reduced with using nonnegative matrix factorization (NMF), data imbalance addressed with the synthetic minority oversampling method (SMOTE), and outliers removed using density-based spatial clustering of applications with noise (DBSCAN). Predictions were made with the extreme gradient boosting (XG Boost) algorithm. Key factors associated with asthma included symptoms like dry cough, runny nose, breathing difficulty, and tiredness. The results can help clinicians predict asthma onset early and support timely intervention. Results: According to experimental findings, XG Boost classifier approach provided the most accurate results. Our model achieved 99.62% accuracy and area under the curve (AUC) of 0.992. Conclusions: This study investigates ML methods for predicting asthma onset in children, identifying XG Boost as the most accurate classifier.