Your Night's Watch: Leveraging Mi-Band-3 Smartwatches and Machine Learning for Detecting Nocturnal Asthma Attacks

This study integrates machine learning and wearable technology to detect nocturnal asthma attacks from daily questionnaire responses, patient information, and Mi-Band-3 smartwatch data. Four models-Logistic Regression, Naive Bayes, Random Forest, and XGBoost-were trained and tuned using grid search and five-fold cross-validation. XGBoost achieved the best performance on the test set, with an AUC of 0.87, an AUPRC of 0.71, and a 2.55-fold increase in precision. Models combining active- (patient-reported) and passive-(device-collected) monitoring features outperformed passive-monitoring only approaches, emphasizing the importance of multiple data sources. Key predictors included an indication of asthma trigger encounter, maximum expected PEF, age, obesity, and sleep quality metrics, revealing complex interactions. Although the Mi-Band-3 contributed valuable information, it could not fully replace active monitoring. Future work should incorporate a larger, more diverse participant pool, integrate additional asthma-related variables, and explore advanced time-series models to improve predictive accuracy and reduce patient burden.Clinical Relevance- Machine learning-driven detection modeling can serve as an early warning system for nocturnal asthma attacks, integrating diverse risk factors and wearable sensor data to improve accuracy. Such a tool could shift asthma management towards a more preventative, personalized approach-minimizing reliance on burdensome self-monitoring, enhancing patient quality of life, and enabling clinicians to intervene proactively when nocturnal symptoms are likely to worsen.