The accurate and timely diagnosis of skin diseases is crucial for effective treatment and improved patient outcomes. However, disparities in diagnostic accuracy across skin tones have been noted and identified, particularly affecting people with darker skin tones. These are based on different factors: underrepresentation of varying skin tones in medical education and training, and biases in current diagnostic tools. This paper explores the prospects for deep learning to cross this gap and arrive at an even skin disease diagnosis across diverse skin tones. We develop and evaluate deep learning models through transfer learning techniques from convolutional neural networks pre-trained with the rich and diverse SCIN (Skin Condition Identification Network) dataset. The problem of data insufficiency and class imbalance in our dataset is mitigated using strategic data augmentation techniques. Our experimental results show a significant increase in overall diagnostic accuracy, with the best-performing model scoring 92.1% accuracy on held-out test data. Also, the meticulous analysis of model performance against different skin tones was done using the FST (Fitzpatrick Skin Type) and MST (Monk Skin Tone) scales. The findings suggest that our deep learning-based decision support system for clinical use reduces diagnostic disparities to provide equal accuracy for all skin tones. Additionally, we use visualization techniques like Grad-CAM to provide insights into the model’s decision-making process, hence making the model interpretable and encouraging trust in the AI-driven diagnosis. This work shows strong empirical evidence that deep learning, when combined with a representative and comprehensive data set, holds a lot of promise in contributing to more equitable and precise diagnosis of skin diseases for all patients, irrespective of skin tone.