Binary Classification of Asthma for the CAPS Pediatric Dataset in Malawi Using Machine Learning

Childhood asthma poses a significant public health challenge, especially in low- and middle-income countries. An early intervention is essential for effective management and improved prevention of Childhood asthma. This study aims to develop a predictive model for childhood asthma by applying machine learning (ML) techniques. The dataset includes self-reported information on respiratory symptoms, anthropometric measurements, spirometry data, and personal carbon monoxide (CO) exposure among children aged 6–8 years in rural Malawi. We employed a supervised ML approach, focusing on classification algorithms and handling imbalanced outcomes, including Random Forest, Logistic Regression, and XGBoost. Additionally, this study applied the Synthetic Minority Over-sampling Technique (SMOTE), creating synthetic samples of the minority class to balance the distribution of the outcome variable in the training data. Data preprocessing involved handling missing values, feature selection, and normalization to ensure data quality and model performance. Model evaluation was conducted using cross-validation and performance metrics, including precision, recall, and F1-score. Among the evaluated models, Logistic Regression emerged as the most balanced approach, offering strong precision and the highest F1-score while maintaining a reasonable recall rate. This balance reduces the likelihood of overdiagnosis while still capturing a significant proportion of true positives, making it suitable for early screening applications. Moreover, Logistic regression, with its simple mathematical structure, provides more transparency and explainability, which are vital for clinical adoption and gaining practitioner trust.