Year: 2026
Type: conference-paper
Abstract: Asthma is a major pediatric health concern, and timely and precise diagnosis is crucial for proper management. Traditional statistical models plus early machine learning approaches provided interpretability but were limited in capturing non-linear interactions and handling heterogeneous questionnaire data. This pioneering study, marking the first application of machine learning to detect asthma based on questionnaire data, applies transformer-based architectures—TabTransformer and TabNet—to high-dimensional data from 2,730 children aged 6–7 years in Mysuru, India, obtained from the Phase I survey conducted by the Global Asthma Network (GAN). The dataset comprises 160 features, including 127 categorical and 31 continuous variables. Classical models, including Random Forest (RF) plus Support Vector Machine (SVM), are also assessed for comparison. In a highly imbalanced dataset (1.7% positive cases), TabTransformer achieved the superior results, including Matthews Correlation Coefficient (MCC) and F1-score (0.9655) on the test set, zero false negatives (recall = 1.0000), and high precision (0.9333), outperforming TabNet and classical models. Overall, the outcomes. demonstrate that the transformer-based approach can effectively handle complex, heterogeneous tabular data, providing reliable early-stage asthma detection and highlighting their potential to improve pediatric healthcare outcomes. Our novel approach enables cost-effective asthma screening in resource-constrained rural India, where tests like spirometry, CT scans, or MRIs are impractical due to cost and resources, by identifying high-risk individuals for targeted testing, reducing costs and manpower.
Source: Procedia Computer Science
Language: English
Subfield: Health Information Management
Field: Health Professions
Domain: Health Sciences
Open Access status: diamond
OpenAlex ID: W7167672264