Performance analysis of artificial intelligence-based classification models for diagnosing asthma in children.

INTRODUCTION

Asthma is a common childhood disease with symptoms such as cough, wheezing, and shortness of breath. This study evaluated the role of artificial intelligence in improving diagnostic accuracy in children.

METHODS

We included patients aged 6-18 years evaluated at our clinic between January 2024 and January 2025. Those with chronic cough were classified as asthma or non-asthma based on final diagnosis. Demographic, clinical, and pulmonary function data were collected. Eight machine learning models Gradient Boosting, AdaBoost, Random Forest, Logistic Regression, Linear Discriminant Analysis, Decision Tree, k-Nearest Neighbors, and Naive Bayes were applied, and their performance was assessed using accuracy, precision, recall, F1 score, ROC AUC, and MCC.

RESULTS

A total of 900 children were included, with 450 diagnosed with asthma and 450 with non-asthmatic chronic cough. Males comprised 52.9% of the cohort. Feature importance analysis highlighted exercise-induced cough and recurrent bronchiolitis as the most significant predictors for asthma. Gradient Boosting demonstrated the highest diagnostic performance (F1: 0.974, ROC AUC: 0.997), followed closely by Random Forest (F1: 0.972, ROC AUC: 0.997) and AdaBoost (F1: 0.969, ROC AUC: 0.995). Logistic Regression, LDA, Decision Tree, and Naive Bayes showed moderate performance, while KNN had the lowest accuracy (F1: 0.566, ROC AUC: 0.615), indicating variable effectiveness among models.

DISCUSSION: Machine learning algorithms show promise in improving diagnostic accuracy and efficiency in pediatric asthma, though further research is needed.