An Investigation and Comparative Study of Skin Cancer Classification Using Transfer Learning Techniques

Skin cancer poses a significant health challenge globally, including in India, where the incidence of skin cancer has been rising steadily. Accurate and timely detection is crucial for effective treatment and better patient outcomes. Traditionally, dermatologists diagnose skin cancer through visual inspection, which can be time-consuming and subject to human error. Recent advancements in deep learning algorithms have shown great potential in improving the accuracy and efficiency of skin cancer diagnosis. This study presents a deep learning model for skin cancer prediction using the HAM10000 dataset, a comprehensive collection of skin lesion images with corresponding diagnoses. Our model employs convolutional neural network (CNN) architectures with different activation functions, along with transfer learning models AlexNet and LeNet, which have proven successful in image classification tasks. We assess the performance of our models using metrics such as accuracy, precision, and recall. Our findings indicate that the transfer learning based AlexNet architecture achieves the highest accuracy of 97.59% on our dataset. This research demonstrates that transfer learning models can significantly enhance the accuracy and efficiency of skin cancer diagnosis. It also underscores the importance of training these models with extensive and diverse datasets. Moreover, our results suggest that more complex CNN models, such as AlexNet, may be better suited for skin cancer prediction compared to simpler architectures like LeNet.