Development of Coronary Heart Disease Diagnosis System Based on Facial Imagery

Abstract

The effort of the community in early checkups for coronary heart disease is still very lacking. That is due to the less awareness and cost constraints so that heart disease is handled too late which causes the heart condition to get worse and even complications. A diagnosis of coronary heart risk based on facial texture has been done lately by researchers in China and produces pretty good output, but no similar study has been found in Indonesia. Therefore, this study aims to design several machine learning models and find out the performance of algorithms in diagnosing the risk of coronary heart disease with facial imagery. The research dataset was extracted using the Gray Level Co-occurrence Matrix on specific areas of the face. The areas taken are right crow's feet, right canthus, bridge nose, forehead, left canthus and left crow's feet. The main focus of this study was on the performance of the Support Vector Machine, Decision Tree, and Neural Network models. Dataset processing procedures are divided into two, namely model making (training) and performance testing (testing). The findings showed that the best performing model was NN with an AUC score of 92.8%, followed by SVM with 85.6%, and the lowest was DT with 68% AUC. NN became the model with the best performance with 76.9% accuracy, 81.7% precision, 76.9% recall, and an F1 score of 77.5%.