Deep Learning Dermoscopy: Unveiling CNN-SVM Synergy in Skin Lesion Detection

The created algorithm was assessed rigorously, employing numerous performance metrics that were calculated per the categories of skin lesion detection. The precision for each class ranged from 75.00% to 87.85%, indicating that the majority of the model’s positive predictions were events that did indeed occur. To manage Class 5 to be the best in precision compared to other levels. The recall metrics, which means the announced level of the model’s accuracy in searching for all examples from each class, has a performance range of 72.84% to 84.81%, with Class 2 having the greatest recall. The F1-Score, which includes accuracy and recall, corresponds to a percentage that ranges from 73.90% to 83.00%. Further, the study applied various types of averages in order to establish how the model would apply to different classes. The model workout was done on the wise 79.30%, 79.40%, and 79.23% micro-average for accuracy, recall, and F1-score. Results thus attest that an equal performance in all the classes was achieved without the effect of any class imbalance. The weighted mean, giving similar consideration to the occurrence of real cases in each class, generated accuracy, recall, and F1 scores of 79.48%, 79.24%, and 79.24%, respectively, having a small better bent for classes having greater prevalence. Finally, the Micro-average is demonstrated as it is made up of contributions of all classes, and it shows consistent performance across all those Groups, having accuracy, recalling, and F1-Score all equal to 79.24%. As a result of this research, AI skin lesion identification systems are exposed to the application of CNN-SVM models that can perfectly manage medical diagnostics in future, and the shorter the adaptions and the further model simplification, the higher the relevance of performance evaluation options.