Melanoma is the most aggressive form of skin cancer responsible for a significant percentage of skin cancer-related fatalities. However, existing deep learning-based melanoma detection methods face several challenges, including dataset imbalance, which affects model generalizability across different skin tones. Additionally, many models require extensive computational resources, lack interpretability, and struggle with variations in image quality and lesion morphology, leading to potential misclassifications. To overcome these challenges, a novel deep learning-based approaches for automated melanoma detection using Random Forest. Initially, the input image is preprocessed using Gaussian filters (GF) to enhance clarity by reducing noise. The deep learning-based ResNet50 is used for extracting features from the Melanoma images. Finally, the extracted features are subsequently fed into Random Forest (RM) classifier that classify Melanoma and Non-Melanoma. The effectiveness of the proposed AMENET was evaluated using F1 score, accuracy, precision, recall, and specificity. The proposed AMENET model achieved a classification accuracy of 98.88%. The proposed AMENET model enhances the total accuracy by 3.85%, 0.86%, 1.03% better than ConvNeXtV2, DSCIMABNet, CNN-ViT respectively.