An Efficient Deep Learning based Ensemble Approach for Multiclass Skin Cancer Classification Using Medical Imaging

Skin cancer is considered one of the most common types of cancer that generally happens due to the abnormal growth of skin cells. There are different types of skin cancer with varied symptoms. Various reasons can be there for skin cancer. Any type of cancer can be deadly and harmful. To avoid the risk, the disease requires to be detected at the earliest stage. Identifying the disease is the very first step of the diagnosis process. Any delay in diagnosis may greatly threaten the patient’s life. To identify the disease, different modern techniques are being used among which deep learning shows remarkable performance in classification tasks. Various deep learning-based algorithms have been suggested and built by multiple researchers to categorize the disease. But most of the existing systems have some significant gaps and limitations regarding datasets, data preparation, model selection and development. To remove those gaps and to make the detection process more accurate and reliable, a novel ensemble-based deep learning framework has been proposed. Through this work, multiclass class classification has been performed to detect several types of skin cancer. Four CNN models have been trained separately with proper optimization and finally, a weighted average ensemble has been done to enhance the overall performance. Data preprocessing with hyperparameter tuning reduces issues like overfitting for achieving good results.