Skin cancer, a widespread and sometimes lethal ailment, originates from the anomalous proliferation of skin cells. Prolonged exposure to ultraviolet (UV) radiation often causes it. Early detection and treatment can lead to potential cures. Recently, artificial intelligence (AI)-powered methods for diagnosing skin cancer have shown improved precision and effectiveness. Traditional methods reliant on visual inspection by dermatologists may be error-prone, especially for small or challenging lesions. Deep learning (DL) models have proven effective in analyzing extensive image datasets and extracting pertinent features for classification. Pairwise correlation-based representations, like covariance matrix-based visual representations, have shown excellent performance in image classification in DL models. However, confounding effects can disrupt these representations as they identify pairwise correlations between feature components. Addressing this concern, this paper employs a deep visual representation method centered on partial correlation. For skin cancer image classification, this paper uses the iterative sparse inverse covariance estimation (iSICE) algorithm along with VGG-19, which features batch normalization. The experimental findings indicate the efficacy of the proposed method for categorizing skin lesion images into several classes. It attains a 93.74% accuracy on the HAM10000 dataset.