Skin cancer is one of the most common types of cancer, and early detection is critical for effective treatment. This work compares three deep learning models for melanoma classification: a simple CNN, a CNN with Dropout, and MobileNetV2 with transfer learning. All were trained on a public dataset of 10,000 images resized to 64×64 pixels. While this resolution improved training speed, it likely hindered MobileNetV2’s performance due to architectural constraints. We evaluated accuracy, precision, recall, and F1-score. Results show that lightweight CNN outperformed MobileNetV2, emphasizing the need to align input resolution with model architecture.