Random Forest for Data Aggregation to Monitor and Predict COVID-19 Using Edge Networks

Abstract

Medical sensors and distributed edge networks hold promise in advanced control and prediction of infectious diseases, such as COVID-19. Their integration can lower communication latency, bandwidth utilization, and energy consumption while achieving high application scalability and reliability. Existing edge-devices-enabled e-healthcare frameworks face two serious challenges that affect their success. First, real-time medical sensors generate similar or redundant readings used for disease prediction, mostly for non-COVID-19 patients, which increases data transmission latency and energy consumption. Second, predicting the risk level of COVID-19 patients using lightweight machine learning requires minimal training time to be useful. To address these challenges, we develop an edge-centric e-healthcare framework for online health data monitoring and analysis to predict the risk level of COVID-19 patients. In the proposed framework, a statistical data aggregation method (MEAN function) is deployed at local gateway devices to remove redundant data to minimize communication latency and energy usage. Geo-distributed edge servers are used to predict with 97 percent accuracy the risk level to each patient with Random Forest (RF) on the aggregated data. Finally, the efficiency of the RF algorithm is demonstrated with standard classification techniques.