Predictive Modeling System for Automated Skin Lesion Classification Using Deep Neural Networks and Voting Ensembles

Ushaa Eswaran; Vishal Eswaran

Abstract

Skin cancer is one of the most prevalent cancers globally. Early and accurate diagnosis is critical for timely treatment and improved prognosis. This study presents a predictive modeling system for automated classification of skin lesions from dermoscopic images using deep neural networks and voting ensemble techniques. A customized 16-layer convolutional neural network architecture is developed for feature learning from lesion images. The concept of horizontal voting ensemble is implemented by training multiple models from different epochs and combining their predictions. The proposed approach is evaluated on a dataset of 3000 dermoscopic images classified into melanoma, seborrheic keratoses, and benign nevi. Results demonstrate that the predictive modeling system with voting ensemble achieves 96.8% sensitivity and 97.2% specificity in differentiating malignant melanoma from benign skin lesions. The system holds promise for improving efficiency and consistency of skin cancer screening.

Abstract

Keywords