Nowadays, skin cancer is a serious health risk to people, and early detection is becoming more and more important. Employing recent studies in artificial intelligence and deep learning this paper constructs a comprehensive automated pipeline using a dataset containing over 300,000 lesions captured by 3D total body photographic modification. The set is based on telemedicine settings by Incorp- rating important metadata with high-quality images of the lesions, including age, sex, and anatomical regions. The finest classification performance was achieved with Efficient Net V1B0 and Edge Next deep learning models. Efficient Net V1B0 was able to outperform many popular models on this task by combining image-based classification with machine learning based on metadata. A greater degree of customization was achieved by employing more complicated feature engineering aimed at creating new indices that quantify lesion characteristics. In prediction, the Voting Classifier ensemble, which was created using the advantages of the various algorithms, performed best in efficiency. The ensemble outperformed its single model counterparts by virtue of improved accuracy which the experimental results present enabled the ensemble to receive high ROC AUC on telehealth and clinical images from diverse lesion types. In conclusion, this study proposes using deep learning alongside other machine learning algorithms to tackle the accuracy deficit in skin cancer diagnosis, making it easy to make accurate diagnoses remotely.