Skip to main content

ORIGINAL RESEARCH article

Front. Big Data
Sec. Medicine and Public Health
Volume 8 - 2025 | doi: 10.3389/fdata.2025.1503883
This article is part of the Research Topic Deep Transfer Learning in Public Health: Opportunities for Innovation and Improvement View all articles

Next-Generation Approach to Skin Disorder Prediction Employing Hybrid Deep Transfer Learning

Provisionally accepted
Yonis Gulzar Yonis Gulzar 1*Shivani Agarwal Shivani Agarwal 2Meenakshi Kandpal Meenakshi Kandpal 3Sherzod Turaev Sherzod Turaev 4*Choo W Onn Choo W Onn 5Shilpa Saini Shilpa Saini 6Abdenour Bounsiar Abdenour Bounsiar 1
  • 1 King Faisal University, Al-Ahsa, Saudi Arabia
  • 2 Ajay Kumar Garg Engineering College, Ghaziabad, India
  • 3 Odhisa University of Technology and Research, Bhubaneswar, Odisha, India
  • 4 United Arab Emirates University, Al-Ain, Abu Dhabi, United Arab Emirates
  • 5 INTI International University, Nilai, Negeri Sembilan Darul Khusus, Malaysia
  • 6 Chandigarh University, Mohali, Punjab, India

The final, formatted version of the article will be published soon.

    Skin diseases significantly impact an individual's health and mental well-being, yet their classification poses challenges due to complexity, overlapping symptoms, and varying lesion characteristics. Traditional convolutional neural networks (CNNs) often struggle with generalization when trained on limited datasets, leading to suboptimal classification results. In this research, we propose a hybrid model that integrates DenseNet121 and EfficientNetB0 through a Hybrid Deep Transfer Learning Method (HDTLM) to enhance model generalization in skin disease prediction. DenseNet121 excels at capturing intricate patterns through its dense connectivity, while EfficientNetB0 provides computational efficiency and scalability. This model was rigorously trained on a diverse dataset of 19 skin conditions, consisting of 19171 images, achieving an impressive training accuracy of 98.18% and a validation accuracy of 97.57%. We conducted a comprehensive comparison with state-of-the-art models, including DenseNet121, EfficientNetB0, VGG19, MobileNetV2, and AlexNet, demonstrating that our hybrid approach consistently outperforms these models across multiple evaluation metrics, including precision (0.95), recall (0.96), F1-score (0.95), and accuracy (98.18%). The results underscore the effectiveness of the HDTLM in addressing the challenges of skin disease classification, providing a robust solution for tasks characterized by significant domain shifts and limited labeled data.

    Keywords: skin disorder prediction, deep learning, Transfer Learning, DenseNet121, EfficientNetB0, Computer Vision, image classification

    Received: 29 Sep 2024; Accepted: 03 Feb 2025.

    Copyright: © 2025 Gulzar, Agarwal, Kandpal, Turaev, Onn, Saini and Bounsiar. This is an open-access article distributed under the terms of the Creative Commons Attribution License (CC BY). The use, distribution or reproduction in other forums is permitted, provided the original author(s) or licensor are credited and that the original publication in this journal is cited, in accordance with accepted academic practice. No use, distribution or reproduction is permitted which does not comply with these terms.

    * Correspondence:
    Yonis Gulzar, King Faisal University, Al-Ahsa, Saudi Arabia
    Sherzod Turaev, United Arab Emirates University, Al-Ain, Abu Dhabi, United Arab Emirates

    Disclaimer: All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article or claim that may be made by its manufacturer is not guaranteed or endorsed by the publisher.