Improved stacked ensemble with genetic algorithm for automatic ECG diagnosis of children living in high-altitude areas

•For the first time, we conducted a study on the automatic diagnosis of pediatric ECGs in high-altitude areas.•We developed a model that combines a GA and the stacked ensemble method, capable of differentiating between normal and abnormal ECGs and classify four types of rhythm.•We incorporated not only popular ML models but also the most successful deep learning models for tabular data tasks in recent years into the baseline models, enabling us to demonstrate the advantages of the developed model.•We conducted a characteristic importance analysis, revealing important factors contributing to ECG abnormalities in children living in high-altitude regions. Electrocardiogram (ECG) is a commonly used diagnostic tool in clinical practice that plays a vital role in the diagnosis and treatment of various heart diseases. Previous studies have indicated that the incidence of heart disease in children living in high-altitude areas is significantly higher than those in low-altitude areas, which increases the demand for ECG examinations in these regions. However, there is a lack of research and technology that focus on automatic ECG diagnosis for children in high-altitude areas. This study utilized electronic medical records of pediatric electrocardiograms (ECGs) from high-altitude areas in Yunnan Province, China as research data. An improved genetic algorithm (GA) was employed to find the best combination of base classifiers, and a stacked ensemble approach was utilized to develop a reliable model for automatic diagnosis of pediatric ECGs. The developed model is capable of identifying abnormal ECGs and classifying heart rhythm types into four categories: normal sinus rhythm, sinus tachycardia, sinus bradycardia, and other arrhythmias. The model was tested and compared with commonly used classification methods. The results show that the model developed in this paper exhibits better performance in terms of accuracy, recall, and F1 score. The model’s classification effectiveness was further demonstrated by creating a receiver operating characteristic (ROC) curve. Finally, feature importance analysis highlighted the significance heart rate, developmental stage, and QRS complex axis deviation angle in the model.