"A Hybrid CatBoost–NGBoost Model for Predicting Asthma Attack Risk and Severity from Environmental Triggers"

Abstract

Asthma is a chronic respiratory condition that is highly sensitive to environmental changes such as air pollution, humidity, temperature, and allergen levels. Early prediction of asthma attacks using environmental data can play a critical role in preventing life-threatening episodes and improving patient quality of life. This study explores the application of machine learning techniques—CatBoost and NGBoost—to predict the likelihood and severity of asthma attacks triggered by environmental factors. CatBoost, a gradient boosting algorithm optimized for categorical features, is used to classify the risk of an asthma attack based on real-time environmental inputs such as air quality index (AQI), pollen levels, humidity, and temperature. Its ability to handle categorical data and missing values makes it ideal for health-related tabular datasets. In parallel, NGBoost (Natural Gradient Boosting) is employed to model the uncertainty and probabilistic distribution of asthma attack severity, providing not just point estimates but also confidence intervals for better risk assessment. The integration of these models enables a robust and interpretable asthma risk prediction system capable of supporting real-time alerts for at-risk individuals. The results demonstrate high predictive accuracy with CatBoost and meaningful probabilistic forecasts using NGBoost, validating their effectiveness in clinical and environmental health monitoring scenarios. This approach offers a promising step toward AI-assisted asthma management, potentially empowering both patients and healthcare providers with timely, data-driven insights.