Leveraging Efficient Net for Deep Learning-Driven Hazardous Chemicals Induced Skin Diseases Classification

Abstract

Skin disease classification plays a crucial role in enhancing diagnostic processes and treatment planning in dermatology. In this research, a Deep Learning model, built upon the EfficientNet framework highlights admirable capability in categorizing six conventional skin diseases due to exposure of hazardous chemicals on skin such as Atopic Dermatitis, Discoid Eczema, Dyshidrotic Eczema, Contact Dermatitis, Neurodermatitis and Seborrheic Dermatitis. The model exhibits a balanced learning approach, reflected in its high training accuracy (84.59%) and commendable validation accuracy (66.82%). The minimal gap between training and validation accuracies suggests a successful avoidance of overfitting, emphasizing the model's potential for real-world applications.

Performance analysis reveals specific strengths and areas for improvement in the model's classification metrics. Notably, Atopic Dermatitis showcases balanced precision (0.482) and sensitivity (0.647), resulting in an F1 Score of 0.552. However, Contact Dermatitis calls for improvement in achieving a better balance between precision and sensitivity. The model's performance on unseen images is highly impressive, achieving 100% accuracy in identifying all six skin disorders induced by exposure to hazardous chemicals.

The EfficientB0 model exhibits a mix of strengths and weaknesses in its performance metrics, with notable proficiency in specificity but room for improvement in precision, recall, and overall accuracy. While it effectively identifies negative instances, its ability to capture positive instances and achieve a balanced classification is suboptimal. This abstract underscores the need for optimization to enhance the model's precision-recall balance and overall classification accuracy.Top of Form

In conclusion, this research presents a robust Deep Learning model for skin disease classification, offering valuable insights into its learning capabilities and areas for refinement. The model's high accuracy on unseen data underscores its potential utility in clinical settings, contributing to enhanced and reliable skin disorder diagnoses. Future work may focus on refining specific classes to further elevate the model's performance and expand its applicability in diverse dermatological scenarios due to exposure of hazardous chemicals on skin.