Melanoma is the most threatening type of skin cancer due to its rapid spread and the high mortality rates associated with it. Early and accurate diagnosis relies on advanced techniques, such as deep learning, for classifying skin lesions. The aim of this study is to develop an intelligent model for the early detection and classification of melanoma using deep learning techniques. This is achieved by utilizing a high-resolution dataset of melanoma skin images. The study follows a unique methodology to achieve optimal accuracy by employing a simplified pre-processing approach based on edge detection and selective fine-tuning of the final layers of the InceptionV3 model. This methodology reduces the reliance on heavy data augmentation and uses techniques to prevent overfitting, resulting in a more generalized model. The study realized strong results in generalizing melanoma classification, with an accuracy of 97% and a high ability to distinguish between benign and malignant lesions. The results indicate that the model quickly learned the distinguishing features for melanoma classification with minimal overfitting.