Dyah Darma Andayani, Andi Aisyah Nurfitri, Fhatiah Adiba, Muhammad Yahya
Melanoma, a highly perilous form of skin cancer, is triggered by exposure to ultraviolet light, leading to DNA (deoxyribonucleic acid) damage in skin cells. Often, melanoma resembles an ordinary mole, making it challenging to differentiate. Hence, there is a pressing need for an early detection system capable of discerning melanoma from non-melanoma skin lesions. This study employed the reinforcement learning technique with the Deep Q-Network (DQN) algorithm model. The process encompassed dataset collection, preprocessing, sharing, DQN model development, training, and evaluation. The dataset was categorized into melanoma and non-melanoma classes, comprising 2,625 and 2,538 skin lesion images, respectively. The outcomes revealed that employing 16x16 pixel images and 300,000 timesteps yielded a remarkable accuracy of 80.62%, achieved within a mere 841 seconds. This early detection system aims to enhance melanoma diagnosis, boosting cure rates and minimizing its impact. Additionally, this research contributes significantly to advancing skin cancer detection technology. © 2024 Author(s).
Departement of Computer Engineering, State University of Makassar, Makassar, Indonesia; Departement of Automotive Engineering Education, State University of Makassar, Makassar, Indonesia
