Automating Melanoma Detection for Early Diagnosis using Conventional Machine Learning and Deep Learning Techniques

Melanoma poses a significant global health threat, where early and accurate detection is crucial for improving patient survival rates.However, the visual diagnosis of skin lesions is often subjective and challenging due to the high similarity between benign moles and early-stage melanoma.This paper addresses this challenge by developing and evaluating a robust automated system for distinguishing between benign moles and melanoma using machine learning and deep learning techniques.A key aspect of the methodology was a hybrid feature engineering approach, combining clinically inspired ABCDE rule metrics with textural features from Local Binary Patterns (LBP) and color statistics.Several classification models were systematically evaluated, including traditional machine learning algorithms (Support Vector Machine, K-Nearest Neighbors, and Random Forest) and deep learning architectures (MobileNetV2 and AlexNet) on the Melanoma Skin Cancer Dataset.The experimental results demonstrated the superiority of the proposed AlexNet model over all other models tested, which achieved an outstanding classification accuracy of 95.2% and an Area Under the ROC Curve (AUC) of 0.99.Further, the contribution of this paper is extended to a practical application.The "Smart Skin Analyzer" desktop application and a "Melanoma Detector" Android application were developed to translate the research into a tangible tool, aiming to achieve the goal of raising awareness and facilitating early melanoma detection.