Skin diseases are disorders that affect the normal structure or function of the skin, leading to irregularities in skin such as rashes, inflammation, discoloration, lesions, itching, pain, or changes in skin texture. In India, skin diseases are one of the major health problems due to the number of factors like tropical climate, large population, diverse skin tones and limited access to dermatological care. Although the majority of Indian skin diseases are not life-threatening, they can still make daily life incredibly uncomfortable and stressful. The skin disease classifiers are dominated by lighter skin tones (Fitzpatrick Types I–III), and research shows that there is an accuracy drop of 30-40\% when these models are tested on darker skin tones. As a result, these models are less reliable for populations with darker skin tones, including the Indian population. Since Indian skin tones predominantly belong to Fitzpatrick Types IV–VI, there is currently no such model that can correctly classify Indian skin diseases. So there is a need for a model that can correctly classify Indian skin diseases. In this paper, we present a model that uses a two-stage fusion-based approach. In the first stage, the model predicts one of 8 main-class disease groups (achieving 68.19% Top-1 and 93.91\% Top-3 accuracy), and the second stage predicts one of 19 sub-class diseases, reaching up to 84.15% accuracy. 

dermatology AI,skin disease classification,Indian skin tones,EfficientNetV2,metadata fusion,fairness analysis