Embedded Flu Detection System based Cough Sound using MFCC and kNN Algorithm

Flu is an infectious respiratory disease caused by viruses that affected people every year. In some cases, flu can lead to hospitalization and even death. Figure out whether the flu also affects community is considered challenging for preventing wide outbreaks. Therefore, we propose an efficient flu detection using embedded system. The proposed method for flu detection based on Mel Frequency Cepstral Coefficient (MFCC) and K-Nearest Neighbor (kNN) algorithm which can effectively distinguish cough sounds. The cough is a flu symptom since flu will cause a dry cough. The effectiveness of the proposed method is evaluated in the task of classifying the flu or non-flu from the 270 subjects of the COUGHVID dataset that provides crowdsourced cough recordings. Furthermore, the proposed method is implementing on embedded devices-based Raspberry Pi. Based on experimental results, the proposed system presents 80% of accuracy, 4.370 seconds of execution time, and 27% of Central Processing Unit (CPU) usage. This result indicates that the proposed system is non-invasive, reliable, easy to apply, and can reduce the workload testing in local health centers.