MOBILE APPLICATION FOR COUGH SOUND PROCESSING FOR RESPIRATORY DISEASE PRE-SCREENING

• Main Author: Mohammad, Zawilhikam
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/87728
• Institution: Institut Teknologi Bandung

Respiratory diseases are one of the hot topics discussed worldwide since the emergence of COVID-19. The risk of spreading diseases that attack the respiratory tract is generally very high. In 2022, there were 633 million cases with deaths reaching 6.6 million people. Lack of awareness and knowledge of the symptoms experienced is one of the factors in the very rapid spread. There are many cases where the host of a virus does not realize that he is carrying the virus and has the potential to spread it. It is necessary to develop a tool to carry out pre-screening by utilizing the Intelligent Engineering Platform (PRC) as a framework for thinking in software development and utilizing microservices architecture to ensure that the applications built are able to handle scalability challenges. The research was conducted using the Design Research Methodology (DRM) method with the initial step of conducting Research Clarification (RC) regarding user identification and application limitations. Users will be divided into three, namely users, hospitals and admins. The limitation of the application to be built is that it can only be operated on Android Mobile devices. The application that is built will not replace the function of medical personnel because its nature is only to carry out pre-screening. Next, Descriptive Study I (DS-I) reviews related literature as Knowledge material to be applied. There are several previous related studies so that application development only focuses on integrating into the Android application being built. There are two relevant models, namely the Singularity Characterization (SC) and K-Nearest Neighbor (KNN) models that will be applied. The next step is to carry out a Prescriptive Study (PS) by developing by paying attention to the rules in developing Machine Learning (ML) applications, namely incoming cough sound data will be preprocessed with a segmentation of only one second and an assumption of 48,000 samples per second. Furthermore, the sound will go through the noise reduction stage with noisereduce.py in python. Sounds that have gone through the preprocessing stage will go to the feature extraction stage using the SC model so that they get three parameters, namely dimension, size and dispersion. The data will be processed separately by two models, namely the SC model and the KNN model. The classification results will be compared with AND logic so that the final results can be displayed to the user. The last stage is Descriptive Study II (DS-II), which is to evaluate the results of application development. The first evaluation to check the algorithm applied by comparing the results when only using the SC model 59.7% and the SC-KNN model 56.7%. Based on these results, the combination of the SC-KNN model has not been able to improve the accuracy of detecting the COVID-19 respiratory disease. The second test is on the ability of the application built to face scalability challenges. Testing using Apache JMeter HTTP Request with a user limit of 1000 resulted in the application being able to handle up to 5107 requests which periodically increased from 79 users to 1000 users per request. The server time in handling requests is relatively fast, which is an average of 0.7 seconds. This study has succeeded in implementing PRC as a framework for thinking using the PSKVE-S (Product, Service, Knowledge, Value, Environment and System) architecture. The position of this research in PSKVE-S focuses on implementing Knowledge from Medical and IT Experts into a Product by considering Server, Process, Service Cost, Energy and Cost from previous related research.