SKIN CANCER CLASSIFICATION AND DETECTION USING MODIFIED EFFICIENTNETB7

Abstract

Cancer is one of the most lethal diseases that threatens the humanity today. Skin cancer is one of the most dangerous of cancer. Skin cancer rates 6^th most type of the cancer that are increasing globally.  Patients' chances of survival are very low when they get inadequate therapy and inaccurate diagnosis. The patient's prognosis and likelihood of survival improve with each passing day that the illness is detected. As a result, early diagnosis and treatment may be challenging and intricate. To tackle these problems, we need a technology that can diagnose and identify skin cancer sooner on its own. To achieve the best results from the CNN model, we started by preparing the publicly available dataset. This involved improving the photos’ colour, contrast, and clarity, and then resizing and fine-tuning the data. We tested various CNN models with these enhanced datasets, and found that EfficientNetB7 performed the best, achieving a score of 86.81%. To further improve performance, we made additional adjustments to the EfficientNetB7 model by adding more layers and tweaking hyperparameters. We evaluated the performance of the final model using several metrics, including sensitivity, specificity, misclassification rate, and accuracy. The model achieved an accuracy of95.98%, a specificity of 85.4%, a sensitivity of 86.38% and misclassification rate of 9.39%. This led to our suggested approach surpassing state-of-the-art CNN designs and techniques.