Electronic Theses and Dissertation

Penyakit kulit menular seperti monkeypox, chickenpox, measles, dan hand-foot-mouth disease (hfmd) memiliki kemiripan gejala pada lesi kulit sehingga berpotensi menimbulkan kesalahan identifikasi pada tahap awal. penelitian ini bertujuan membangun model klasifikasi penyakit kulit menular berbasis yolo, membandingkan performa yolov5s dan yolov8s, serta menerapkan model terbaik ke dalam prototype aplikasi berbasis android. dataset penelitian disusun melalui integrasi tiga sumber data, yaitu monkeypox and similar skin disease dataset, msid, dan msld v2.0, sehingga menghasilkan lima kelas, yaitu monkeypox, chickenpox, measles, hfmd, dan normal. tahap pra-pemrosesan meliputi pembersihan data, pembagian subset, anotasi gambar, integrasi dataset, serta penyeimbangan data latih melalui oversampling berbasis augmentasi. pelatihan dilakukan pada lima konfigurasi dataset eksperimen dengan variasi image size, jumlah epoch, learning rate, dan batch size. evaluasi model dilakukan menggunakan metrik accuracy, precision, recall, f1-score, map50, dan inference time. hasil penelitian menunjukkan bahwa pada dataset integrasi, skema penyeimbangan dengan target 1747 citra per kelas memberikan hasil yang lebih seimbang dibandingkan target 3357 citra per kelas, dan yolov8s memberikan performa yang lebih baik dibandingkan yolov5s secara keseluruhan. model terbaik diperoleh pada yolov8s konfigurasi d3, yaitu konfigurasi dataset integrasi dengan penyeimbangan data latih hingga 1747 citra per kelas, menggunakan image size 640, 150 epoch, batch size 32, dan learning rate 0.001. model ini menghasilkan accuracy 0,862, precision 0,792, recall 0,455, f1-score 0,578, map50 0,532, dan inference time 4,0 ms. model terbaik tersebut kemudian diterapkan ke dalam prototype aplikasi android untuk mendukung pengujian awal hasil klasifikasi penyakit kulit menular.

Infectious skin diseases such as monkeypox, chickenpox, measles, and hand-foot-mouth disease (HFMD) have similar skin lesion symptoms that may lead to misidentification at an early stage. This study aims to develop a YOLO-based classification model for infectious skin diseases, compare the performance of YOLOv5s and YOLOv8s, and implement the best model into an Android-based prototype application. The dataset was constructed by integrating the Monkeypox and Similar Skin Disease dataset, MSID, and MSLD v2.0 into five classes: monkeypox, chickenpox, measles, HFMD, and normal. Preprocessing included data cleaning, subset splitting, image annotation, dataset integration, and augmentation-based oversampling for training data balancing. Training was conducted on five experimental dataset configurations with variations in image size, epochs, learning rate, and batch size. Evaluation used accuracy, precision, recall, F1-score, mAP50, and inference time. The results show that, on the integrated dataset, balancing with a target of 1747 images per class produced more balanced results than 3357 images per class, and YOLOv8s outperformed YOLOv5s overall. The best model was YOLOv8s on configuration d3 with an image size of 640, 150 epochs, a batch size of 32, and a learning rate of 0.001. It achieved 0.862 accuracy, 0.792 precision, 0.455 recall, 0.578 F1-score, 0.532 mAP50, and 4.0 ms inference time. The model was then implemented into an Android-based prototype application for initial testing of infectious skin disease classification results.