A skin cancer diagnostic classifier is introduced in this thesis. Comparisons are made with several models, showing ways to solve imabalance data in clinical data. In addition, the importance of calibrating the classification models is implemented and explained. Bayesian neural networks are implemented that allows the incorporation of other factors that influence skin cancer. Finally, it is discussed that a diagnosis with the help of CNN with Bayesian NN can be beneficial and more accurate, improving the diagnostic accuracy of doctors in the detection of melanoma.
Datasets: Dataset 2019, Dataset 2020 & Resized dataset
Thesis (in Spanish): Document
Power point presentation of thesis: Presentation