Detection of Skin Diseases and Vitamin Deficiency-Related Conditions Using Machine Learning and Deep Learning

This research proposes the use of advanced deep learning architecture referred to as EfficientN et for diagnos ing various skin conditions, ranging from Acne, to Rosacea, Pigmentary Disorders, Eczema, and Psoriasis, using simple images captured with any cameras. The methodology involves the four critical stages: image labeling, preprocessing, extracting features, and classification. Median filtering is used for image pre processing to remove noise, hair and other artifacts while pre serving the essential edges, textures, and optimal image analysis. Feature extraction uses techniques such as Shearlet Transforms (ST) and Blur Trace (BT) to capture the characteristics of skin texture, including curvature and energy distribution. These extracted features are fed into the EfficientN et model, which classifies the skin condition into predefined classes, which are Acne, Rosacea, Pigmentation Disorders, Eczema, and Psoriasis. Classification results of EfficientN et were observed to be high with well supported metrics such as precision, recall, and F1 score. Cross-validation assures robustness and adaptability across diverse datasets. Unlike previous methods based on dermoscopy images, simple images from digital cameras broaden the system applicability and make it suitable for real-world applications. This work aims to develop an accurate, scalable diagnostic tool that will help healthcare professionals with early detection and personalized treatment of dermatological conditions associated with vitamin deficiencies. The dataset that has been used is a public dataset available on Kaggle.