Analisis Perbandingan Kinerja Model Yolov7 dalam Deteksi Kuku Diabetes

Abstract

Abstract

               Diabetes mellitus (DM) is a degenerative and non-communicable disease that can be seen from the color of the fingernails. In analyzing color the human eye has limitations in color recognition and texture analysis while computers are able to classify millions of colors and slight texture changes to recognize changes in individual nail color to prevent early symptoms of diabetes using the YOLOv7 method to represent a one-stage model for detecting objects using a Convolutional Neural Network ( CNN).

               This research was carried out at the Polewali Community Health Center. Sampling was carried out by taking medical records and conducting interviews with the relevant doctors. Sample data was taken from several diabetes mellitus patients and several workers at the Polewali Community Health Center for healthy nail sample data.

               The results of testing the YOLOv7 model with epoch 100 showed accuracy of 81%, precision of 82.4%, recall of 95.5% and F1-Score of 88.5%. Testing the YOLOv7 model with epoch 200 resulted in an accuracy of 90%, precision of 93.3%, recall of 93.3% and F1-Score of 93.3%. Testing the YOLOv7-x model with epoch 100 resulted in an accuracy of 71.4%, precision 72.3%, recall 82.9% and F1-Score 77.2%. Testing the YOLOv7-x model with epoch 200 resulted in an accuracy of 63.3%, precision 60.4%, recall 90.6% and F1-Score 72.5%. Testing the YOLOv7-tiny model with epoch 100 resulted in an accuracy of 91.4%, precision 95.6%, recall 93.5% and F1-Score 94.5%. Testing the YOLOv7-tiny model with epoch 200 resulted in an accuracy of 94.6%, precision 93%, recall 100% and F1-Score 96.4%. The results of comparative testing of the YOLOv7 model in detecting diabetic nails, concluded that the ideal model that can be used is the YOLOv7-tiny model with an epoch value of 200.

Keywords: Confusion Matrix, CNN, Diabetes Mellitus, Nails, YOLOv7