Optimizing Acne Diagnosis: A CNN-SVM Hybrid Framework for Accurate Classification of Acne Lesions

A good study on the classification of acne focused on three types including nodules, blackheads, and papules. The results were excellent on all categories of the model; regarding nodules, precision, recall, and F1-score, each was 96.30%. This means the model was always consistent and accurate in terms of classification for nodules. Blackheads presented a slightly decreased level of performance. Precision, recall, and F1-score reached $96.25 \%$, while still showing exceptional results. The papules showed the best performance with $96.39 \%$ precision, recall, and F1-score. The support values for each class were 4050 for nodules, 4000 for blackheads, and 4150 for papules. The figures indicate that the dataset is balanced well, with a support proportion of 0.33 for both nodules and blackheads and 0.34 for papules. This distribution ensures that each acne type is represented fairly in the analysis. Besides that, the model scored an excellent $\mathbf{0. 9 8}$ for all three acne classes, showing the model to be reliable and sound. There are 12,200 images in the dataset. In general, the model shows $96.31 \%$ accuracy overall. On average, the model has high precision and recall capabilities and thus good predictive powers that deliver precise and consistent classification results for types of acne. All such results indicate that these models are suited for some kinds of clinical applicability: Acne detection grading can be easily automated for its efficient treatment processes. Balanced performance in the classes indicates model robustness towards real-world realizations.