Emerging Soft Computation Tools for Skin Cancer Diagnostics

A non-invasive technique for skin cancer diagnostics that reliably classifies lesions as malignant or benign is analyzed and preferred using machine-learning and deep-learning algorithms. The different stages of diagnostics involve using machine learning: a collection of data images, filtering the images to remove unwanted details and noise, segmenting the images using various clustering algorithms. Feature extraction methods have been used to accomplish classification. Five distinctive classifiers have been trained and their efficiency has been compared. K-nearest neighbor, support vector machine, decision trees, multi-layer perceptron, and random forest are used to classify the skin lesion as malignant or benign. An effective comparison of two different deep-learning architectures, such as AlexNet and GoogLeNet, has been carried out. The dataset, which contains 900 images, is subjected to various identification techniques. The accuracy, F-measure precision, and recall were used to evaluate the effectiveness of the classification scheme. As a result of the findings, when compared with the random forest classifier, AlexNet has a high accuracy of 95%. The number of training samples seems to have a serious influence on the ability of deep-learning strategies. Although having a small number of training samples, the presented scheme was able to precisely discriminate among healthy and diseased lesions. Hence, the proposed method will enhance the effectiveness of early detection for skin cancer and could be used in computer-assisted systems to help dermatologists discover cancerous lesions.