Skin cancer remains one of the most common and deadly forms of cancer globally, requiring precise and effective diagnostic tools. This study introduces a novel approach for improving skin cancer detection and classification by incorporating Genetic Algorithm Optimized Ensemble (GAOE) methods with transfer learning and meta-learning techniques. We refined three well-known pre-trained models—EfficientNetB3, ResNet50, and VGG16 using the ISIC 2020 dataset. We used transfer learning, customizing them for skin cancer detection employing a weighted ensemble approach, where initial performance-based weights were calculated and optimized via evolutionary algorithms. The GAOE model, which integrates genetic algorithms and meta-learning, performed exceptional performance accuracy of 99.43%. The GAOE model also surpassed the weighted ensemble in precision 98%, recall 99%, and F-score 99%. Combining ensemble approaches, optimization algorithms, and transfer learning, our methodology efficiently addresses typical problems, including underfitting, overfitting, and model bias. This study provides a strong tool for skin cancer diagnosis and a precedent for future research in medical imaging, highlighting the possibility of incorporating evolutionary algorithms and meta-learning to enhance predictive performance.