Developing Improved Melanoma Detection Strategies Using Hybrid CNN and Autoencoder Models and Detailed Data Analysis

• DOI: 10.52783/jisem.v10i9s.1238
• Publication Date: 2025-02-10T10:32:32Z
• Authors: Vinay S. Nalawade

Melanoma is a very dangerous type of skin cancer that needs new ways to be found so that it can be treated quickly. The goal of this study is to create better ways to find melanoma by combining mixed Convolutional Neural Networks (CNN) and Autoencoder models with in-depth data analysis. This will make melanoma identification more accurate and reliable. The chosen method takes advantage of deep learning techniques that are designed to work well with image-based classification tasks. This study used a dataset from the Skin Cancer MNIST: HAM10000 for this work. It has dermatoscopic pictures that need to be carefully preprocessed. To make the models work, this meant shrinking, normalising pixel values, one-hot encoding, and rearranging the pictures. Data enrichment methods like rotations, shifts, and flips were used to make the model even better by training it against a wider range of data representations. In the methods part, we describe how we set up our hybrid model, which blends the feature extraction skills of CNNs with the dimensionality reduction skills of Autoencoders. With settings like an epoch count of 25 and a batch size of 128, this setup went through a lot of training using the Adam optimiser and Categorical Cross Entropy as the loss function. Our results show that the mixed model works, as it achieved a trial accuracy of 73.19% with a loss of 1.079. There are more details in the classification report and confusion matrices that show how well the model works with different types of skin lesions. For example, it is much better at finding arterial lesions, with a precision of 66% and a recall of 100%. Comparing the mixed model to normal CNN models shows that it is more accurate and precise; this proves that it is better at diagnosing problems. This study shows that mixing CNNs and Autoencoders could be useful for making accurate screening tools for cancer. This could lead to future improvements in medical imaging technology.