Enhancing Local Feature Detection Performance with Sequential CNN Architecture in Keras

Early diagnosis and classification of skin lesions may improve prognosis for cancer prevention. The skin lesion dataset pictures provide a number of important difficulties with regard to the characteristics that may be used to create strong and adaptable cross-domain classification models. This research attempted to pick relevant characteristics from skin lesion datasets for effective skin cancer classification by measuring the model’s performance. This work is primarily concerned with preprocessing the HAM10000 skin lesion dataset in order to identify essential variables that will drive effective skin cancer categorization. The study employs K-fold cross-validation for testing and goes through the process of creating the training model with CNN adopting a sequential architecture. We suggested K-fold cross-validation as an alternative to numerous multiple approaches to minimise the complexity. It has been noted that the K-fold approach on the dataset demonstrates the maximum accuracy performance around 97.88%. In contrast with existing methods, the proposed approach finally yields the best accuracy results.