<p>Skin cancer detection is a vital research area due to the high incidence of the disease and the critical need for early diagnosis to improve survival rates. While traditional methods have primarily focused on image-based analysis of skin lesions using machine learning (ML), diagnosing skin diseases from clinical images remains one of the most challenging tasks in medical image analysis. This study introduces a hybrid learning-based approach that integrates both image features and clinical data to enhance diagnostic accuracy. By incorporating patient clinical data, the proposed model provides a more comprehensive analysis of skin cancer cases. A significant challenge is the limited availability of labeled data, particularly for minority lesion types. To address this problem, a Conditional Generative Adversarial Network (cGAN) is employed to generate synthetic data and reduce class imbalance. The hybrid model leverages both high-level learned features and low-level engineered features to improve classification accuracy. Experimental results demonstrate that the proposed learning model-based CNN residual learning and Lasso regularization for feature selection achieved an accuracy of 0.968, an AUC of 0.950, and a loss of 0.008, outperforming traditional methods and showing improved performance in accurately classifying underrepresented lesion types. Moreover, the proposed model was compared with different embedded feature selection methods, achieving high accuracy (Lasso: 0.968, Ridge: 0.962, and Elastic Net: 0.959). The results show that the proposed model is robust across diverse patient demographics and lesion types, highlighting its potential for clinical application. The findings indicate that integrating clinical data with image-based analysis offers a powerful tool for improving diagnostic accuracy in skin cancer detection, particularly for early diagnosis.</p>