Skin Cancer Classification Utilizing a Hybrid Model of Machine Learning Models Trained on Dermoscopic Images

Skin cancer is the most prevalent kind of cancer, and early recognition is critical for successful treatment. Computer-aided skin cancer detection systems have been developed to use image analysis techniques to examine skin lesions and detect suspicious ones. Artificial intelligence techniques, especially machine learning, have been widely applied for skin cancer detection. Using machine learning algorithms, it is possible to scrutinize vast databases of skin lesion images, recognize patterns, and classify lesions as either benign or malignant. Similarly, machine learning algorithms can learn to identify key features from skin lesion images and classify them with remarkable precision. Technology integration has transformed the area of skin cancer detection, and computer-assisted methods have the potential to increase the speed and accuracy of skin cancer diagnosis, resulting in better patient outcomes. Our model uses several machine learning algorithms proposed to classify skin cancer, such as Convolutional Neural Networks (CNNs), Support Vector Machine (SVM), Naïve Bayes, and K-Nearest Neighbor (K-NN) classifiers. The research problem revolves around developing algorithms and techniques to analyze skin lesion images and accurately categorize them as either benign or malignant. The primary aim is to develop an accurate, robust, and automated system that can assist clinicians in the early-stage diagnosis of skin cancer, thereby enhancing patient outcomes and reducing healthcare costs. This paper evaluates various machine learning algorithms and selects a model based on its optimal performance in predicting and classifying skin cancer. The dataset utilized in this paper for training and testing the model comprises 1439 images of benign skin lesions and 1196 images of melanoma lesions, gathered from the International Skin Image Collaboration (ISIC). The proposed model is composed of five steps: image acquisition, preprocessing, segmentation, feature extraction, and classification. In addition to using machine learning algorithms for evaluating the model, such as CNNS, SVM, Naïve Bayes, and (K-NN) classifiers, we achieved 99.5%, 95.6%, 86.1% and 77.1%, respectively. Furthermore, our results demonstrate that Convolutional Neural Networks (CNNs) performed most accurately in visual image tasks, as shown in our proposed work.