Deep Learning-Based Classification of Skin Lesion Dermoscopic Images for Melanoma Diagnosis

Abstract

This study examined the effectiveness of artificial intelligence and machine learning methods in accurately categorizing skin cancers. We used the ISIC 2019 Skin Lesion dataset, which consists of images divided into eight categories. To prepare the data, we applied preprocessing techniques and extracted features using the SqueezeNet deep learning model. The dataset was divided into training and test sets using cross validation. Four well-known machine learning algorithms, namely Artificial Neural Network (ANN), k-nearest neighbors (kNN), Random Forest (RF), and Logistic Regression (LR), were employed to perform classification tasks. Each algorithm was specifically designed to suit its processing methodology. The algorithms are assessed based on several important measures. The results indicate that the Artificial Neural Network (ANN) achieved the highest accuracy rate of 71.80%, whereas the k-nearest neighbors (kNN), Random Forest (RF), and Logistic Regression (LR) achieved accuracy rates of 69.70%, 67.00%, and 67.20%, respectively. The results emphasize the capacity of machine learning algorithms to augment clinical decision-making in the field of dermatology, with the goal of enhancing early detection and treatment effectiveness for individuals with skin cancer.This study examined the effectiveness of artificial intelligence and machine learning methods in accurately categorizing skin cancers. We used the ISIC 2019 Skin Lesion dataset, which consists of images divided into eight categories. To prepare the data, we applied preprocessing techniques and extracted features using the SqueezeNet deep learning model. The dataset was divided into training and test sets using cross validation. Four well-known machine learning algorithms, namely Artificial Neural Network (ANN), k-nearest neighbors (kNN), Random Forest (RF), and Logistic Regression (LR), were employed to perform classification tasks. Each algorithm was specifically designed to suit its processing methodology. The algorithms are assessed based on several important measures. The results indicate that the Artificial Neural Network (ANN) achieved the highest accuracy rate of 71.80%, whereas the k-nearest neighbors (kNN), Random Forest (RF), and Logistic Regression (LR) achieved accuracy rates of 69.70%, 67.00%, and 67.20%, respectively. The results emphasize the capacity of machine learning algorithms to augment clinical decision-making in the field of dermatology, with the goal of enhancing early detection and treatment effectiveness for individuals with skin cancer.