Automated Skin Disease Classification Using Deep Neural Networks: A Study on a Bangladeshi Dataset

Abstract

Skin diseases are a significant public health concern in Bangladesh, with a high prevalence among the population. Accurate and timely diagnosis is essential for effective treatment, but challenges such as limited access to dermatologists in rural areas and the high cost of medical consultations persist. This study proposes a deep learning-based framework for the automated detection of skin diseases in Bangladeshi people, leveraging advanced image processing techniques. Six state-of-the-art deep neural network architectures, DenseNet201, InceptionV3, MobileNet, NASNetLarge, VGG19, and Xception, were trained and evaluated using a dataset curated specifically for the Bangladeshi population. The dataset consisted of dermatologically annotated skin disease images that were preprocessed and augmented to enhance model generalization. Performance evaluation was conducted based on accuracy, precision, recall, and F1 score. Among the tested architectures, NASNetLarge achieved the highest classification accuracy of 91%, demonstrating its potential for reliable skin disease detection. This research provides a robust solution to assist dermatologists, improve healthcare accessibility, and optimize resource allocation in Bangladesh. The developed framework has the potential to be integrated into mobile and web applications for real-time disease detection, thus improving early diagnosis and patient outcomes across the country.