An ensemble asthma prediction model for early detection of asthma in children using stacking and blending

Asthma disease is a serious worldwide health challenge affecting every age bracket, particularly amongst children. Its widespread has extended to numerous countries. Childhood asthma remains underdiagnosed and insufficiently treated in Nigeria, affecting more than 20,000 children, including adults. It was discovered that there is a notable incidence of wheezing, with an approximation of 13 million or more individuals suffering from asthma. The objective of this study is to build an ensemble asthma prediction model for early detection of asthma in children using stacking and blending. The methodology involved combining several techniques, including Relief feature selection, grid search optimization, and a blended stacking ensemble model. This study used the Nigerian dataset comprising the Hospital Record System (HRS) of six Nigerian hospitals and the administrative data from patients' history. The findings indicate that the proposed model attained a notably high precision of 96%, a recall rate of 94%, and an F1-score of 95% for the non-asthmatic category (0). Furthermore, it achieved a precision of 96%, a recall of 97%, and an F1-score of 97% for the asthmatic positive category (1). The study concludes that the model demonstrates effective capability in distinguishing between asthmatic and non-asthmatic patients. Consequently, this could contribute to improved patient outcomes and enhanced healthcare delivery systems. This work proposes further exploration, such as external validation, incorporating more diverse datasets and enhancing interpretability or explainability of the model.