Comprehensive Skin Disease Identification Using Augmented Data and Transfer Learning: Achieving Optimal Accuracy with Sequential Models

Skin diseases are a significant global concern, and effectively treating bacterial and fungal skin infections poses an enormous task. Furthermore, in more severe instances, it might occasionally result in a case of skin cancer. Diagnosing skin diseases from clinical photographs is challenging in medical image analysis. In addition, the process of diagnosing skin issues manually by medical professionals is both time-consuming and subjective. In this work, Deep learning and machine learning-based model has been designed for the identification of skin disease using the concept of transfer learning. For this, the VGG16 model, sequential, KNN, logistic regression, and decision tree classifiers are used. VGG16's basic architecture comprises 16 weight layers, with 13 convolutional and 3 completely linked. Because ability to accurately extract complex features from skin photos is a result of its simplicity and depth. The model's ability to generalize is enhanced by its pretrained training on huge image datasets such as ImageNet. Excellent performance with little medical data can be achieved by fine-tuning this pretrained model on specific skin disease datasets. Improving patient outcomes can be achieved through earlier diagnosis and treatment, which is made possible by its ability to detect subtle patterns and early indicators of skin illnesses that may not be visible to the human eye. These proposed models are evaluated on three datasets obtained from Kaggle. These datasets now include 246 of our own captured images, which are added as healthy classes. Data augmentation techniques are used to improve the randomness of the input dataset in order to ensure model stability. The sequential model works best with an overall accuracy of 96% with the Adam optimizer and 10 epochs, outperforming state-of-the-art techniques. This model will help dermatologists diagnose skin diseases early.