Automated Skin Tone Classification Using InceptionV3: Enhancing Accuracy and Inclusivity in Image Recognition

The study describes the skin tone classification into four categories, such as dark, light, mid-dark, and mid-light, using the machine learning of InceptionV3.The objective of this work was to improve the accuracy and fairness of automatic skin tone classification. The model has achieved 820 overall accuracy when trained and tested on the dataset containing 7,166 images. Performance metrics such as precision, recall, and fl-score were used to understand how this model performs across diverse skin tones. The results are that the best classes from the model are “dark” and “mid-dark”, with precisions, recalls, and fl-scores of 0.91, 0.92, and 0.92 and 0.92, 0.79 and 0.85, respectively. While for the class “light,” the performance was somewhat lower, with a precision of 0.79 and recall of 0.87, thus giving an fl-score of 0.83, the lowest metrics concerned the “mid-light” class, which had a precision of 0.63, recall of 0.66, and an fl-score of 0.64. Both the macro-averaged and weighted-average scores of the metrics were 0.81 and 0.82, respectively, which underlines the model's effectiveness balance across skin tones. In fact, the study emphasizes the promise of the InceptionV3in the classification of skin tones and points to some shortcomings, most especially the light skin tone. This work has to be improved in the future by enhancing model performance toward balanced skin tone classification, handling biases of the model, and expansion of the dataset for applicative usefulness in real world scenarios.