PUTRI, Fayza Nayla Riyana and Isnanto, Rizal and Sugiharto, Aris (2025) OPTIMASI HYPERPARAMETER DENSENET DENGAN RANDOM SEARCH DAN PARTICLE SWARM OPTIMIZATION UNTUK KLASIFIKASI JENIS RUAM KULIT MANUSIA. Masters thesis, UNIVERSITAS DIPONEGORO.

	Text COVER.pdf Restricted to Repository staff only Download (1MB)
	Text BAB I .pdf Download (293kB)
	Text BAB II.pdf Download (3MB)
	Text BAB III.pdf Restricted to Repository staff only Download (724kB)
	Text BAB IV.pdf Restricted to Repository staff only Download (3MB)
	Text BAB V.pdf Restricted to Repository staff only Download (275kB)
	Text DAFTAR PUSTAKA.pdf Download (327kB)

Abstract

Penelitian ini membahas klasifikasi jenis ruam kulit manusia, termasuk cacar air, campak, cacar monyet, cacar sapi, HFMD, dan kulit normal, menggunakan model Deep CNN berbasis DenseNet-201. Fokus utama penelitian ini adalah menentukan kombinasi hyperparameter optimal melalui tiga pendekatan: Random Search, Particle Swarm Optimization (PSO), dan kombinasi Random Search dengan PSO (RS+PSO). Hasil penelitian menunjukkan bahwa metode RS+PSO memberikan performa terbaik dengan kombinasi hyperparameter optimal, yaitu learning rate 0,00009, dropout rate 0,57, batch size 64, filters 687, dan optimizer RMSprop. Evaluasi model menunjukkan akurasi keseluruhan sebesar 93%, dengan nilai ratarata makro untuk precision, recall, dan f1-score masing-masing sebesar 93%, 92%, dan 93%. Model menunjukkan performa tinggi pada setiap kelas, namun terdapat tantangan pada kelas Chickenpox dan Measles dengan nilai recall yang lebih rendah. Setelah memperoleh model terbaik, sistem dikembangkan dengan berbasis Flask yang mampu memproses data secara lokal, menjaga keamanan pengguna, dan menunjukkan potensi besar dalam mendukung deteksi dini ruam kulit secara otomatis.
Kata Kunci : Deep CNN, DenseNet201, Random Search, PSO, Hyperparameter

This study discusses the classification of human skin rash types, including chickenpox, measles, monkeypox, cowpox, HFMD, and normal skin, using a Deep CNN model based on DenseNet-201. The main focus of this research is to determine the optimal hyperparameter combination through three approaches: Random Search, Particle Swarm Optimization (PSO), and a combination of Random Search and PSO (RS+PSO). The results show that the RS+PSO method delivers the best performance with the optimal hyperparameter combination, which includes a learning rate of 0.00009, dropout rate of 0.57, batch size of 64, 687 filters, and RMSprop optimizer. Model evaluation shows an overall accuracy of 93%, with macro-average values for precision, recall, and f1-score of 93%, 92%, and 93%, respectively. The model demonstrates high performance in each class, but challenges remain in the Chickenpox and Measles classes with lower recall values. After obtaining the best model, the system is developed using Flask, which is capable of processing data locally, ensuring user security, and demonstrating significant potential in supporting the automatic early detection of skin rashes.
Keywords : Deep CNN, DenseNet201, Random Search, PSO, Hyperparameter

Actions (login required)

View Item