GlowSkinFit: machine learning and CNN-based approaches for accurate skin disease detection

The diagnosis of skin diseases has been a focus of great interest because of the increased rates of skin disorders and the necessity of prompt and convenient diagnosis of dermatological disorders. In this paper, we introduce a machine-learning model of the multi-class classification of nine common skin diseases with the help of the customized Convolutional Neural Network (CNN). Eight hundred and seventy-eight dermoscopic images were acquired at Kaggle, processed, and augmented followed by classification using the proposed CNN architecture. The images were downsized to 200 × 200 pixels, and rescaled, rotated, sheared, zoomed and horizontally flipped. The model attained a training accuracy of 92.78% and the validation accuracy of 81%. Confusion matrix, precision, recall and F1-score were used in measuring performance. These findings suggest that CNN-based models may be useful in automated scripts of dermatological screening and can be used as the foundation of deployable diagnostic tools.