Research on the automated detection of melanoma from dermoscopy images is swiftly advancing due to the surge in this deadly disease. Melanoma, the most aggressive form of skin cancer, causes 55,500 annual deaths, emphasizing the importance of early identification. However, discerning subtle skin pattern changes is labor-intensive and subjective. Machine learning methods are a powerful tool for visualizing patterns in medical images, enabling swift analysis of large datasets and detection of subtle pattern changes. This study introduces two versions of MobileNet models and the state-of-the-art EfficientNeT B7 for automatic melanoma classification. Concatenating MobileNets V1 and V2 and training end-to-end enhance the model’s learning capacity, leveraging V1’s small footprint and V2’s heightened accuracy. EfficientNeT B7, known for high accuracy and large capacity, further elevates the model’s capabilities. The concatenation of MobileNet V2 with EfficientNeT B7 and end-to-end training improved overall model performance. To address overfitting and achieve generalization, ensembling of the five models was carried out. Prediction results on 10,982 test samples from the SIIM-ISIC melanoma challenge dataset produced AUC-ROC scores of 0.7457, 0.7500, 0.7792, 0.7907, 0.8019, and 0.8169 for MobileNet V1, V2, Concatenation (MobileNet V1 and V2), EfficientNeT B7, Concatenation (MobileNet V2 and EfficientNeT B7), and Ensembling, respectively.