PATTERNED BASED SKIN DISEASE DETECTION WITH OPTIMIZED MACHINE LEARNING MODELS

Abstract

Skin diseases, ranging from common infections to life-threatening conditions such as melanoma, require timely and accurate diagnosis to ensure effective treatment and improved patient outcomes. However, conventional diagnostic methods often rely on dermatological expertise, which may be scarce in remote or resource-constrained regions. Recent advances in machine learning (ML) have enabled automated skin disease detection using dermoscopic and clinical images, showing performance levels comparable to expert dermatologists. Despite these successes, challenges such as high computational complexity, class imbalance, and poor generalization across diverse populations limit the practical deployment of these models. This study explores the use of optimized machine learning models for enhancing skin disease detection. Optimization strategies including transfer learning, hyperparameter tuning, model pruning, quantization, and ensemble methods are employed to improve model accuracy, reduce computational cost, and enable real-time inference on mobile and edge devices. Experimental results demonstrate that optimized models achieve superior classification performance while maintaining low latency and efficiency, thus making them suitable for integration into teledermatology and point-of-care diagnostic systems. The findings highlight the potential of optimized ML models in advancing accessible, reliable, and scalable skin disease detection solutions.