Abstract:
Millions of individuals worldwide suffer from skin illnesses, which pose considerable
health and economic difficulties. Accurate and prompt diagnosis is crucial for
successful treatment and management. Traditional treatments relying on
dermatologists' visual assessments can be time-consuming and prone to mistakes.
This study studies the use of transfer learning with fine-tuned convolutional neural
networks (CNNs) to increase the classification accuracy of skin disorders. Five pretrained models—VGG16, InceptionV3, MobileNetV3, EfficientNetB2, and
EfficientNetB5—were chosen for their demonstrated performance in image
recognition tests. The study's objective is to build a credible skin disease
categorization system utilizing these models. Experimental results indicated that the
EfficientNetB2 model attained the best accuracy at 89.37%, followed closely by
EfficientNetB5 with roughly 87%. The findings emphasize the potential of transfer
learning to transform dermatological diagnostics by delivering a reliable and efficient
method to early diagnosis and treatment, therefore improving patient outcomes and
maximizing healthcare resources. Future research should focus on increasing the
dataset and researching more AI approaches to better the diagnostic capabilities of
these models. This technique will further develop their efficacy and dependability in
clinical applications, opening the path for improved patient care and resource
management.
