Skin Cancer Detection and Intensity Analysis using Deep Learning

Skin disorders impact around 900 million individuals worldwide at any given time, making them one of the most prevalent illnesses affecting human health. Deep Convolutional Neural Networks (DCNNs) has expanded to numerous domains, clearing the door for new research activities, building upon the extraordinary success of DCNNs in Computer Vision applications, including image categorization and object detection. This project endeavors to fine-tune DCNNs previously successful with the ImageNet dataset. The focus is on classifying seven types of skin lesions using the HAM10000 dataset, comprising 10,000 dermatoscopic images. Fine-tuning the entire model with Inception V3, resulting in 85.94% accuracy, DenseNet 201 resulting in 87.42%, and the Ensemble model of both Inception V3 and DenseNet 201 resulting in 85.94% accuracy. Over 20 epochs, Inception V3 marginally surpasses DenseNet 201 in the validation set, while DenseNet 201 excels in the test set, achieving an accuracy of 87.42%.