<p>Skin diseases are common across all ages and are significant sources of infection. Diagnosing skin diseases involves a number of tests. The diagnostic process is laborious, time-consuming, and requires extensive understanding, especially for skin diseases with similar symptoms. Atopic dermatitis and contact dermatitis are two skin diseases with similar symptoms that are often misdiagnosed. These are also two common skin diseases in Vietnam. The disease images used in this study were provided by the Central Hospital of Dermatology. The images were taken using common techniques such as smartphones or cameras and labeled by specialists. To improve the diagnosis of these two inflammatory skin diseases, author proposes a technique that utilizes deep learning features extracted by the VGG-16 model. The K-means clustering technique is then applied to the feature set. Next, a deep convolutional neural network model is trained on each cluster. Finally, author uses a technique that combines the above models to produce the final result. The proposed technique achieves accuracy, recall, and F1-score metrics of 73%, 73%, and 71%, respectively.</p>