Skin cancer is a major global health concern, therefore it has to be detected as early as possible for effective intervention and treatment, especially in the cases of lethal melanoma cancer. Generally, to identify melanoma cancer, dermoscopy image analysis is widely used. It’s an non-invasive skin imaging technique that helps to visualize the features of pigmented skin lesions, which are imperceptible through naked eye examination. However, this method is time-consuming and also prone to operator bias due to similarities between the skin cancers. Hence, the automated dermoscopy image analysis has became a very active research field, and recent advances in deep learning shows that it can be an effective approach to detect the melanoma cancer in early stages. Hence for the robust detection of melanoma cancer we propose and investigate three deep learning models based on InceptionNet-V3, EfficientNet-B7 and MobileNet-V3. Our analysis shows that MobileNet-V3 outperformed the three investigated models by achieving an accuracy of 97.34%. Although, among the three models, InceptionNet-V3 provides least accuracy, nevertheless the majority analysis i.e., out of three models, if two models correctly classify melanoma and non-melanoma, even then the accuracy comes to 97.24%, in which TP and TN is 95.62% and 98.84% respectively. To best of our knowledge, this work is first of its kind to achieve such a remarkable accuracy of more than 97%. Therefore, we can say that our proposed robust models can be a reliable and precise diagnostic tool for medical professionals for the detection of melanoma cancer.