Melanoma, a potentially fatal form of skin cancer, represents a global health concern, necessitating efficient and precise detection for improved patient outcomes. However, the subjective and daunting nature of manual examination and diagnosing skin lesions by dermatologists often leads to delayed identification or misdiagnosis. This research paper introduces a novel technique for automated melanoma classification to address this pressing problem. The proposed research methodology combines deep features with rotation-invariant handcrafted local-binary pattern (RI-LBP) features for skin lesion classification. RI-LBP extracts microstructures present at the local level whereas deep features capture the high-level intuitive features present in a skin lesion image. After combining RI-LBP with the deep features, the combined feature vector is used to train and test the XGBoost machine learning classifier for classification purposes. Moreover, the investigation undertakes a comparative analysis of numerous machine learning algorithms to evaluate their efficacy in the classification of melanoma using the suggested combined feature set. The XGBoost classifier has obtained an Area Under the Curve value of 0.91 in differentiating pernicious melanomas and benign skin lesion dermoscopic images. Thus, the proposed method is suitable for aiding dermatologists, enhancing the efficacy and precision of melanoma identification and treatment, and ultimately yielding superior patient care and outcomes.