Abstract:
This study refers to an advanced convolutional neural network (CNN) approach for
detecting skin diseases, with a focus on distinguishing between cancerous and noncancerous conditions. Using Kaggle's 'Melanoma Skin Cancer Dataset', we addressed class
imbalance with rigorous data augmentation, resulting in a balanced dataset of 5000 images
per class. Our proposed multilayer CNN architecture was designed and trained using this
balanced dataset, with the goal of achieving high disease classification accuracy. To
evaluate our approach, we compared our custom CNN architecture to four popular
pretrained models: VGG16, ResNet101, InceptionV3, and MobileNetV2. After extensive
experimentation and evaluation, we discovered that MobileNetV2 consistently
outperformed all other models, achieving an impressive 98.95% accuracy. This result
highlights the effectiveness of MobileNetV2 in accurately detecting skin diseases. This
result shows MobileNetV2's effectiveness in accurately detecting skin diseases, which
outperforms even our proposed CNN architecture. These findings focus on the importance
of selecting the right model architecture for skin disease detection tasks. The superior
performance of MobileNetV2 shows its suitability for real-world applications requiring
accurate and efficient disease diagnosis. Overall, this study adds valuable insights to the
development and evaluation of CNN-based approaches for skin disease detection, with
implications for improving diagnostic accuracy and patient outcomes.
