Convolutional Neural Networks (CNNs) are highly effective in various computer vision tasks. However, their performance is heavily dependent on the quality and quantity of the training data. This study aims to optimise CNN performance using pattern augmentation techniques to address the problem of limited training data in medical image analysis. Using the DenseNet architecture and the HAM10000 dataset of melanoma and non-melanoma images, the researchers applied a range of augmentation methods, including random rotation, brightness adjustment, random zoom, horizontal flip, height and width shift, and random cropping. The methodology involved evaluating the impact of these augmentation techniques on CNN training. The results showed that random crop and horizontal flip significantly improved classification accuracy by 1–2% over 30 epochs. Extending training to 50 epochs further enhanced performance, though not all methods showed significant improvements; random crop and horizontal flip remained the most effective. These findings highlight the crucial role of selecting appropriate augmentation techniques and training epochs in optimizing CNN performance. The study underscores the importance of dataset diversity and augmentation strategies, particularly in medical image analysis where early and accurate diagnosis is critical. Future research should explore different network architectures and datasets to validate and generalize these results.