Towards Simulating a Global Robust Model for Early Asthma Detection

Abstract

Asthma is a chronic non-communicable disease that affects the lungs and can cause breathlessness leading to fatal exacerbation. This disease mainly starts developing in childhood and can affect the lungs and lifestyle throughout life. Ignoring asthma at any age can be fatal. Therefore, this disease should be detected as early as possible. So, in this regard, we propose a machine learning model to predict early asthma in children. We simulated the federated learning process to build the model and created four virtual hospitals. We have simulated federated learning to build a global robust model where multiple datasets from the different institutions can take part in the training process, which can be used in various regions of the world for predicting early asthma. We have trained the models using both IID (Independent and Identically Distributed) and non-IID approaches of splitting the dataset. We also checked the performance of the models by measuring the predictive accuracy and AUC (Area Under the Receiver Operating Characteristic Curve) score for test data. We got a predictive accuracy of 91.57% and 93.68% for the IID and non-IID approaches. At the same time, we got the AUC score of 0.895 and 0.918 for the IID approach and non-IID approaches.