Predicting the Risk of Asthma Development in Youth Using Machine Learning Models (Preprint)

BACKGROUND Asthma is a chronic respiratory disease characterized by wheezing and difficulty breathing, which disproportionally affects 4.7 million children in the U.S. Currently, there is a lack of asthma predictive models for youth with good performance. OBJECTIVE This study aims to build machine learning models to better predict asthma development in youth using easily accessible national survey data. METHODS We analyzed cross-sectional combined 2021 and 2022 National Health Interview Survey (NHIS) data from 9,716 youth subjects with their corresponding parent information. We built several machine learning models with various sampling techniques (under- or over-sampling) for asthma prediction in youth, including XGBoost, Neural Networks, Random Forest, Support Vector Machine (SVM), and Logistic Regression. We examined the associations of potential risk factors identified from both Random Forest and Least Absolute Shrinkage and Selection Operator (LASSO) with asthma in youth. RESULTS Between the different sampling techniques, undersampling the major class (subjects without asthma) yielded the best results in terms of the area under the curve (AUC) and F1 scores for the different predictive models. The Logistic Regression performed the best with the under-sampled data, yielding an AUC score of 0.7654 and an F1 score of 0.3452. In addition, we have identified additional important factors associated with asthma development in youth, such as low family poverty ratio and parents ever had asthma. CONCLUSIONS This study successfully built machine learning models to predict asthma development in youth with good model performance. This will be important for early screening and detection of asthma in youth.