IMPLEMENTASI CONVOLUTIONAL NEURAL NETWORK MODEL UNTUK KLASIFIKASI CITRA LESI KULIT MONKEYPOX

Monkeypox is a contagious skin disease with symptoms similar to those of other skin diseases. Inadequate medical personnel and facilities, especially in remote areas, hinder early detection. To improve the early identification process, this study proposes using a Convolutional Neural Network (CNN) model to automatically classify monkeypox skin lesion images. The model is expected to accurately distinguish between monkeypox and non-monkeypox lesions. This study involved building and comparing several CNN architectures without using a pre-trained model. The model architectures were constructed with varying numbers of Conv2D and MaxPooling2D layers, as well as different dropout values. All images were converted to 224×224 pixels and normalized. The model was trained using ReLU and Softmax activation functions and dropout and L2 regularization techniques to prevent overfitting. Evaluation was performed by comparing the accuracy and loss values of each model. The results showed that the model with three Conv2D and MaxPooling2D layers, as well as dropout, produced the best performance, achieving an accuracy of 91.11% and a loss value of 0.2265. Although the improvement in accuracy is still insignificant and has not surpassed the MobileNetV2 model, the model has major advantages in terms of execution time and resource requirements. It can be trained in just 3 to 4 hours on a T4 GPU and does not require the powerful resources necessary for large pre-trained models, such as VGG16, ResNet50, or Xception. Therefore, this model is competitive with pre-trained models in terms of execution time and resource requirements