Making accurate diagnoses concerning skin diseases is difficult because of the subjective visual inspections that medical professionals conduct to diagnose them. Though many deep learning models for diagnosing skin diseases exist, many research works still report poor classification results for several skin diseases simultaneously. To overcome this limitation, this paper proposes to study an ensemble learning approach by combining several powerful models for improved results through accurate diagnoses. Six of the most powerful models, DenseNet121, VGG16, ResNet18, ResNet50, ViT-B/16, and EfficientNetB3, will first be tested independently for establishing baselines. On the basis of these baselines, two approaches to ensemble learning would be developed. In the first case, the soft voting combination of ViT-B/16 and ResNet50 reports 87.86 % accuracy, with precision of 88.02 % and 87.86 % recall, which is a relative improvement of 2.43 % over the best-performing individual model. In the second case, the integration of the best three models, including ViT-B/16, ResNet50, and DenseNet121, reports 88.11 % accuracy, reflecting a relative improvement of 3.40 %. From these results, it is clear that combining different models is beneficial for improved classification outcomes by considering diverse representations. This method presents a resilient avenue for automated multiclass skin disease recognition.