Skin cancer is the most prevalent and perilous kind of cancer in human beings. Among the various types of dermatological malignancy, melanomas are particularly malignant and responsible for a significant number of cancer-related deaths. Early skin cancer detection plays a crucial role in reducing mortality rates and saving lives. So, Computer-Aided Diagnosis (CAD) systems that are driven by machine learning algorithms can help to detect melanoma early. In this article, we propose an innovative approach to melanoma recognition through the development of a fully automatic CAD system. To elevate the overall quality of input dermatoscopic images, we apply a series of preprocessing techniques such as median filtering and bottom-hat filtering. Besides that, an adaptive segmentation method based on the well-known Otsu thresholding technique is conducted to accurately extract suspected skin lesion regions from the improved input image. Then, we use the Local Binary Pattern (LBP) feature extraction method to characterize segmented skin lesions. This technique enables us to capture relevant information from the lesions effectively. Ultimately, the extracted features are inserted into a Decision Tree (DT) classifier to categorize each melanocytic cutaneous lesion in a given dermatoscopic image as either benign or melanoma. The proposed method is effectively tested and verified using a 10-fold cross-validation approach, achieving 90.35%, 88.47%, and 86.28% for average diagnostic accuracy, sensitivity, and specificity, respectively. The experimentation is conducted on the ISIC database, which contains suspect melanoma skin cancer cases, utilizing the MATLAB environment.