Three of the most common early symptoms found in COVID-19 patients are fever, sore throat and running nose. These symptoms have become the public’s indicators in identifying anyone who is potentially infected by the disease.

The proposed health prediction system takes facial images as input and predicts whether the person in the image is healthy or ill with fever, sore throat or running nose. This project is developed by using traditional machine learning approach. Four feature extraction methods combined with four machine learning classifiers are experimented with and evaluated to find the best model to be integrated into the health prediction system user interface. The feature extraction methods used are Local Binary Pattern (LBP), Principal Component Analysis (PCA), Linear Discriminant Analysis (LDA), and Gabor filter. The classifiers used are Support Vector Machine (SVM), Neural Network (NN), k-Nearest Neighbours (KNN), and Random Forest (RF).

The experimental results of all 16 variants are shown in the tables below.

The best overall model chosen for the health prediction system user interface is the LBP+NN model, with the highest average testing accuracy of 76.84% in the second-level classification. It also performed considerably well in the first-level classification with lesser overfitting than the other models with similar performances, as it obtained 94.38% and 86.87% of average training and testing accuracies, respectively.

Note: This project was developed on Google Colaboratory with Google Drive mounting. The code of the directories of the folders have to be changed accordingly in order for you to run the experiments.