A diagnostic framework combining the Swin Transformer architecture with feature selection methods and a Random Forest classifier has achieved a 98.31 percent diagnostic accuracy when classifying skin diseases, marking a 4.44 percent improvement compared to previous models. The automated multi-class system also reached 98.31 percent recall, 98.36 percent precision, and an F1-score of 98.33 percent when evaluated on original image datasets.

The study addresses critical diagnostic hurdles posed by dermatological conditions such as Monkeypox, Chickenpox, Measles, and other skin diseases. These conditions pose significant challenges for accurate and timely diagnosis due to their visual similarity and variability in lesion presentation.

To assist in the early and precise detection of conditions like Monkeypox, computer-aided diagnosis systems leveraging artificial intelligence have emerged as valuable tools. These systems enable improved healthcare outcomes.

At the center of the newly proposed approach is the Swin Transformer, which operates as the framework’s primary feature extractor. The model leverages the transformer architecture’s capacity to identify and capture intricate visual patterns embedded within high-dimensional dermatological image data.

To refine the extracted representations, the researchers implemented multiple feature selection methods, including Principal Component Analysis, Mutual Information, Chi-square statistical testing, and Recursive Feature Elimination. These techniques were applied to isolate the most relevant visual attributes, thereby reducing data noise and decreasing computational overhead.

The resulting feature sets were evaluated across several machine learning classifiers, including Support Vector Machines, Extreme Gradient Boosting, Naive Bayes, and Random Forest. Using 5-fold cross-validation on original images, combining Chi-square and Recursive Feature Elimination with a Random Forest classifier consistently produced the highest accuracy, confirming the efficiency of paired feature extraction and selection for automated disease diagnosis.

The research was published in the journal Neural Computing and Applications, appearing in Volume 38 as article number 856 in 2026. The work is authored by Jihad A. Qadir and is classified as an original article. The publication date is recorded as 02 November 2026.

Subject of Research: Automated skin disease diagnosis using Swin Transformer and machine learning feature selection

Article Title: A deep framework for Monkeypox disease diagnosis using Swin-transformer and feature selection techniques

Article References: A. Qadir, J. (2026). A deep framework for Monkeypox disease diagnosis using Swin-transformer and feature selection techniques. Neural Computing and Applications, 38(21), Article 856. https://doi.org/10.1007/s00521-026-12508-y

Image Credits: AI Generated

DOI: 10.1007/s00521-026-12508-y

Keywords: Monkeypox, Swin Transformer, Skin Disease Diagnosis, Machine Learning, Feature Selection, Random Forest, Computer-Aided Diagnosis, Dermatology, deep, framework, disease, diagnosis