Evidence record · UDINUS · SDGs
- Unit
- Rektorat UDINUS
- Period
- 2025
- Goals
- Record
- 1b6fef91-a308-488c-a829-82c80d0de486
- Published
- 2026-10-30
Universitas Dian Nuswantoro conducted a research activity to improve the precision of Convolutional Neural Networks (CNNs) for identifying skin cancer. The activity involved Ervina, Shafira Ezza; Shidik, Guruh Fajar; and Purwanto as participants. The research addresses skin cancer as a significant global health issue requiring precise and prompt identification to improve patient results. It also addresses the challenge of finding appropriate CNN parameters and examines hyperparameter optimization as a method for improving CNN performance.
The activity employed the Grey Wolf Optimizer (GWO), a metaheuristic approach, to optimize two CNN hyperparameters: dropout rate and learning rate. The HAM10000 dataset, obtained from Kaggle, was used for training and evaluation. It contains dermatoscopic images classified into seven skin lesion categories: Actinic keratoses, Basal cell carcinoma, Benign keratosis-like lesions, Dermatofibroma, Melanocytic nevi, Melanoma, and Vascular lesions. The research also used three data augmentation methods: rotation, flipping, and noise addition, alongside the combination of GWO with CNN.
The reported result was an increase in CNN accuracy to 87%, exceeding earlier benchmarks, together with F1-scores for all lesion categories. The results emphasize the capability of GWO to enhance the persuasiveness and flexibility of AI-based skin cancer detection models. In relation to Sustainable Development Goal 3, the activity concerns health research through the development and evaluation of a computational approach for identifying skin cancer.