Early detection of skin abnormalities is essential in lowering fatality rates and ensuring timely treatment. However, currently available techniques encounter limitations such as small datasets, variability within the same lesion classes, imbalanced complex data and minor visual differences. These challenges make it difficult for conventional machine learning to accurately classify multi-class skin cancer abnormalities. To overcome these issues, this paper presents a computational model for the efficient identification of skin abnormalities using an applied deep learning framework. In this study, a Dense Convolutional Networks (DenseNets) model with Spatial Pyramid Pooling (SPP) and active learning is applied for data enrichment and the identification of eight classes of dermoscopic skin cancer images, which are extracted from the International Skin Imaging Collaboration (ISIC) Challenge Datasets 2016 and 2018, respectively. In the proposed system, a data augmentation technique is realized by subjecting the features to the Stochastic Gradient Descent with Warm Restarts (SGDR) model, which reduces the overfitting problem and selects effective model parameters. The performance achieves a recall, precision, and F1-score of 96.2%, 97.8%, and 97.0%, respectively. In addition, the application of the Gradient-weighted Class Activation Mapping (GRAD-CAM) image visualization method guarantees model transparency. This technique supports medical experts in the efficient and early-stage detection of cancerous images.