Design and Evaluation of a Hybrid Model Combining Neural Networks and Decision Trees for the Classification of Atopic Dermatitis

Abstract

 Atopic dermatitis is a chronic skin disorder characterized by recurrent inflammation and clinical features that often resemble those of other dermatological diseases, which makes accurate diagnosis challenging. This study aims to investigate the use of digital skin image analysis as a supportive approach for classifying atopic dermatitis through the development of a classification model and its comparison with an alternative method based on a different feature-processing strategy. The study was conducted using a dataset consisting of 1,169 digital images of skin lesions, which were employed for training and evaluation purposes. A hybrid approach combining visual feature extraction with a decision tree–based classification method was developed and compared with a model based on retraining the feature extraction structure. The results demonstrated that the retrained model achieved superior performance, with a classification accuracy of 92.70%, compared to 86.89% for the hybrid model. Further evaluation showed that the proposed approach achieved a balanced performance in terms of sensitivity and specificity, along with a high area under the receiver operating characteristic curve (AUC = 0.9760), indicating reliable and consistent classification outcomes. The findings suggest that image-based analytical models utilizing retraining strategies can effectively enhance the accuracy of atopic dermatitis classification and provide valuable support for clinical decision-making, without replacing direct medical assessment.