Accurate early detection of skin cancer plays an important role in curing it. Skin cancer detection using machine learning and deep learning has been a great achievement in decreasing human error while diagnosing the disease. The majority of studies are being done on skin cancer image classification. However, this study has been done to test the impact of metadata addition on image classification results of skin cancer. The study was performed on PAD-UFES-20. The dataset is composed of various skin images and their metadata. Metadata includes parameters like age, gender, previous cancer history, diameter etc. The metadata is heavily unfilled. The major challenge is to handle the unfilled data. The study proposes several unfilled metadata handling methods like neural networks and machine learning-based approaches. Also, various approaches have been used to find how metadata knowledge can be integrated with images that can be used to improve classification results. Images were preprocessed by improving their quality and removing the noise. Both multiclass and binary class classification results have been compiled. Images multiclass classification showed an accuracy of 72% while filled metadata multiclass classification showed a good accuracy of 86%. Integrating metadata with images gave an accuracy of 95%. Results show a major increase in image classification results when metadata classification results are fused to it.
https://drive.google.com/file/d/1nY4QzJ8RVP4wtP5AqoQmkHjVljWyRNHs/view
