Early Detection of Melanocytic Nevus Using Ensemble Methods Combining the Predictions of Multiple Models for Convolutional Neural Network Image Classification of Melanocytic Nevus

Abstract:This research focuses on the early detection of Melanocytic Nevus using machine learninganddeeplearning models for classifying skin lesions as benign or malignant. The study employs ensembleclassifiers and customized Convolutional Neural Networks (CNNs) to evaluate their performanceonMelanocytic Nevus images. The performance of four kernel-based ensemble models (linear, polynomial,radial basis function, and sigmoid) is compared, with the linear kernel achieving the highest accuracyof94.74%. Precision, sensitivity, and specificity measures further highlight the linear kernel'sstrongperformance. In addition to ensemble classifiers, three customized CNN models based onResNet50andVGG16 architectures are developed and tested. Model-1 outperforms the other two CNNmodelswithanaccuracy of 82.29%, while Model-3 exhibits the highest sensitivity. Pretrained models like VGG16showmoderate accuracy, with ResNet50 initially improving performance but experiencing a declineduetooverfitting at higher epochs. The results demonstrate that ensemble classifiers, particularlywithlinearkernels, offer reliable performance for melanoma detection, while CNNs, despite their potential, requirecareful tuning to avoid overfitting. The research highlights the importance of combiningensemblemethods with deep learning techniques to improve accuracy in early skin cancer detection. Futureworkmay explore hybrid models, larger datasets, and real-time implementation for clinical use. Keywords — Melanocytic Nevus, ensemble classifiers, Convolutional Neural Networks, skincancerdetection, machine learning, deep learning