Diagnosis of COVID-19 in a person should be done right away to lessen the chances of the virus to be transmitted to others. To do the early diagnosis of the disease, medical laboratory and antigen tests were required, but these were not always accessible or readily available. This study proposes a new method for detecting COVID-19 using Artificial Neural Networks (ANN) by analyzing a person's current symptoms being experienced by the person without requiring laboratory tests. ANN was used dataset used in this study collected from Kaggle which is the COVID-19 presence and symptoms dataset. We used GridSearchCV to execute data balancing procedures as part of data pre-processing and hyperparameter tuning, as well as 10-fold cross validation, to get the best ANN performance possible, and a prediction model was constructed utilizing the optimal configuration. The results suggest that hidden layer sizes of (100,), (50, 100, 50), and (50, 50, 50), relu and tanh activation functions, adam solver, 0.05 and 0.0001 alpha values, and adaptive and constant learning rates were the values that achieved the best algorithm performance. The optimal configuration of the ANN algorithm was used to create a prediction model. The developed prediction model attained 98.84% accuracy, 98.79% specificity, 100% sensitivity, and 98.84% ROC curve. This prediction model can be used to create applications that detect the presence of the COVID-19 disease in real – time manner without requiring laboratory tests.

Authors:
Charlyn Rosales, Bulacan State University, Philippines

About the Presenter(s)
Dr Charlyn Rosales is a University Associate Professor/Senior Lecturer at Bulacan State University in Philippines

See this presentation on the full schedule – Saturday Schedule

Bookmark