Melanoma is one of the deadliest class of skin cancer diseases, which is the most prevalent. The key to effective therapy and better patient outcomes is early identification. Deep learning algorithms have been applied to melanoma diagnosis in recent years, and the findings are encouraging. The ResNet50 model, a popular convolutional neural network design noted for its excellent accuracy in image classification, is the one we suggest using in this paper to identify melanoma. In this paper, we present an approach to improve melanoma detection by employing preprocessing techniques and transfer learning with ResNet50 model. To enhance the quality of the dataset, we implemented several preprocessing steps. These include techniques for removing hair to eradicate unwanted artifacts, and techniques for removing noise from the image to reduce image disturbances. Additionally, we incorporated image enhancement algorithms to further refine the dataset, resulting in improved image quality for subsequent analysis. We test our model against a 3297 dataset of skin lesion photos. Comparing its performance to that of other cutting-edge algorithms. Our findings demonstrate that the ResNet50 model surpasses competing models in terms of precision and high accuracy of 94.27% over 48 epochs making it an effective tool for the early identification of melanoma. The methodology of combining preprocessing techniques and transfer learning with ResNet50 model showcased the effectiveness over other methodologies.