Machine learning methods for automating skin lesion classification

Abstract

The skin serves as the body’s primary protective and most visible layer, making it particularly susceptible to lesions that may cause physical discomfort, psychological distress, and potential health risks. Addressing skin diseases is therefore both a medical necessity and a matter of social sustainability, as untreated conditions can significantly reduce quality of life and limit social participation. This paper presents an artificial intelligence–based approach for distinguishing between Acne and Actinic Keratosis using lightweight image-processing techniques that minimize computational requirements and support resource-efficient healthcare technologies. To address the lack of a unified dataset, lesion images were collected from multiple repositories, standardized in size, and enhanced through grayscale illumination. Local Binary Patterns (LBP) were employed for feature extraction and dimensionality reduction, followed by classification using machine learning models. The results show that the Multilayer Perceptron (MLP) achieved higher accuracy (99.5%) than the Support Vector Machine (SVM) (97.5%). Overall, the proposed approach combines diagnostic accuracy with computational efficiency, contributing to accessible and sustainable healthcare solutions that improve patient outcomes while reducing technological resource demands.