A Comparative Analysis of Machine Learning and Deep Learning Algorithms for Skin Cancer Detection and Classification: Exploring Performance Across Multiple Parameters

Skin cancer is a substantial health concern, with a yearly incidence of 123,000 new cases and melanoma being the most prevalent variant. Identifying skin diseases is becoming increasingly challenging due to visual similarities. Automated skin cancer screening devices have the potential to enhance accuracy in identifying skin cancers. Advancements in technology and processing capacities have enabled the use of machine and deep learning algorithms for skin lesion examination. However, there is a pressing need for expedited diagnostic systems for life-threatening illnesses. This study provides an overview of various methodologies used in diagnosing skin cancer using image analysis and examines lessons and insights from previous studies. Future directions and scope will help address issues in autonomous detection of skin cancer. In a comparative analysis of various machine learning models, different performance metrics were observed. The Support Vector Machines (SVM) yielded an accuracy of 75.23%, with an AUC of 90.22%. While effective in high-dimensional spaces, SVMs may face limitations with non-linear or noisy data. Random Forest, an ensemble method, demonstrated superior performance with an accuracy of 80.12% and an AUC of 93.22%, possibly due to its proficiency in handling non-linear data structures. Neural Networks, with their ability to model intricate relationships, achieved an accuracy of 81.34%. Multi-Layer Perceptron (MLP), a subset of neural networks, slightly varied in performance, indicating potential differences in architecture or training. Convolutional Neural Network (CNN), renowned for image recognition tasks, showcased an impressive accuracy of 93.23%, hinting at the dataset's spatial structures. The Long Short-Term Memory Recurrent Neural Network (LSTM-RNN) emerged as the top performer with an accuracy of 94.344%, suggesting the dataset's potential temporal or sequential patterns. This comparative study underscores the importance of selecting the appropriate model based on the nature and structure of the data.