Skin Disease Classification using CNN Deep Learning Technique

Abstract

Skin diseases rank among the most prevalent health conditions globally, yet their diagnosis remains challenging, primarily due to the intricate interplay of factors such as skin tone, color variations, and hair presence. The diverse manifestations, initial symptom similarities, and uneven distribution of lesion samples further compound the complexity of accurately classifying these disorders. Deep Convolutional Neural Networks (CNNs) have exhibited remarkable potential in improving the precision of skin disease classification. This paper presents a novel approach that involves developing a custom-built CNN architecture from scratch to classify skin diseases with high accuracy. Unlike existing models that rely on pre-trained architectures, our model is designed and trained from the ground up, tailored specifically to the unique characteristics of dermatological image data. The architecture comprises multiple convolutional and pooling layers, followed by dense layers with Batch Normalization and ReLU activation to ensure effective learning and generalization. Comprehensive experiments were conducted to evaluate the performance of the proposed CNN model. Results demonstrate that the model achieves a notable accuracy rate of 84.9%, along with a competitive F1-score, confirming its reliability
in practical applications. Despite the absence of transfer learning or pre-trained weights, the proposed model effectively captures discriminative features necessary for skin disease classification. General Terms Image processing, Custom-built CNN model, ReLU and SELU activation functions, Eczema, Melanocytic Nevi, Melanoma Basal Cell Carcinoma, Benign Keratosis Lesions, Atopic Dermatitis Deep Learning techniques, state-of-the-art model, Dermatological image classification, Convolutional Neural Networks (CNNs), Batch Normalization.