Skin cancer is a significant global health concern, necessitating early diagnosis for effective treatment, especially for vulnerable populations like farmers. This study introduces SkinPalNet, a multi-model ensemble technique integrating InceptionV3, EfficientNetV2, and ResNet50, enhanced with data augmentation, preprocessing, GAP layer, Dropout, L2 regularizers, PReLU, BN layers, and additional Dense layers to address underfitting and overfitting. Evaluated on four datasets, SkinPalNet demonstrated superior performance, particularly on Dataset 4, achieving 99.15% accuracy and a 98.58% F1 score. However, Dataset 2, with over 10,000 images, showed slightly lower performance (96.20% accuracy and 95.65% F1 score), indicating a need for further optimization for large datasets. Future research will focus on improving model efficiency, exploring advanced optimization techniques, and expanding the range of skin cancer datasets. Enhancing model transparency through Grad-CAM++ and Score-CAM will provide intuitive analyses, benefiting patients, doctors, and researchers in medical imaging. This study contributes significantly to advancements in skin cancer detection and medical imaging.