Early and accurate detection of skin cancer is critical for enhancing patient prognosis and saving healthcare resources. In this paper, we introduce a comparative evaluation of machine learning (ML) and deep learning (DL) models for machine vision-assisted skin cancer detection on dermoscopic images. The dataset employed in this work consists of images from the ISIC archive and more than 15,000 labeled images of different lesion types. We would like to thank Dr. Hooman Bakhtiari for providing the dataset. Basis models used Random Forest, Support Vector Machine (SVM), Artificial Neural Network (ANN), Gradient Boosting, and a self-designed Convolutional Neural Network (CNN) from scratch. Hybrid preprocessing methods, i.e., data cleaning, normalization, feature extraction, and label encoding, were used to provide the optimal input data quality. Model performances were measured in terms of accuracy, precision, recall, F1-score, AUC, and loss. Experimental results showed that the CNN model performed best among other models with 97.02% accuracy in classification, outperforming ResNet-50, Random Forest, and SVM. The research assures the effectiveness of non-canonical CNN architectures in real dermatological diagnosis. The article presents a comprehensive performance evaluation of existing ML and DL methods in dermoscopic image classification and envisages future prospect in mobile and telehealth usage.