Early detection of melanoma is crucial due to its increasing prevalence, emphasizing the importance of accurate delineation and classification of skin lesions in the dermatological examination. Our study introduces a novel method for both segmenting and classifying skin lesions, employing segmentation for improved feature extraction and posterior lesion detection. Leveraging transfer learning networks such as Inception, SqueezeNet1.0, MobileNetV2, MobileNetV3, DenseNet121, and EfficientNetB0, the learned features are fed into traditional machine learning models to classify the lesions. We evaluate our approach using metrics such as sensitivity, specificity, accuracy, and F1-score to assess its efficiency. Extensive testing on the HAM10000 dataset validates our method’s effectiveness, achieving an accuracy of 98.57%, and improving state-of-the-art models. Furthermore, our parallelization strategy enhances scalability and facilitates seamless integration into high-throughput clinical settings.