Automatic Detection of Acne Types Using the YOLOv5 Method

Abstract

Acne is very common due to several factors such as hormones, hygiene, and environmental exposure. This research aims to develop an automatic detection system for facial skin problems using the You Only Look Once v5 (YOLOv5) algorithm, focusing on the problem of acne types on acne-prone faces, and this research is the latest research that has never been done before. The research methodology was carried out by taking datasets directly on acne faces, with a sample of 1230 images. The research process includes data collection, labeling using the Roboflow platform, dividing the dataset into training, testing, and validation data, and implementing the YOLOv5 algorithm using Google Colab. The research stages include data input, object labeling, dataset configuration, YOLOv5 preparation, modeling, model testing, hyperparameter tuning, and model performance evaluation. The results of this study resulted in an accuracy rate seen based on the mapped value of 87.6%, so this can be considered that the model is considered good in detecting the type of acne on facial skin problems in accordance with testing on data, and this model can be implemented to automatically detect facial skin problems, especially on faces with acne, in the future.