The visual similarity between lesions and the lack of access to expert dermatological examination remains a clinical problem because it is challenging to diagnose with precision and early enough to treat skin diseases. Although the latest developments in artificial intelligence made the image-based diagnostics more efficient, the unimodal systems that only use a dermoscopic image do not necessarily see all the important details of the patients that could be used to improve the diagnostic accuracy. This paper presents a multimodal deep learning system that combines both dermoscopic images of the skin and structured clinical information in order to obtain a more accurate classification of skin diseases. In the approach, HAM10000 dataset is used and it consists of 10,015 annotated dermoscopic images with variables including patient age, sex, and lesion location. A Convolutional Neural Network (CNN) backbone is used to extract image features whereas Multilayer Perceptron (MLP) network is used to encode clinical attributes. They are merged with an attention-directed mechanism to extract complementary information of modalities. The experimental analysis proves that the proposed model attains 94.7% accuracy, F1-score of 0.93 and AUC of 0.96, which is better than image-only and clinical-only baselines. The findings prove the hypothesis that clinical metadata combined with visual features can greatly improve the classification robustness and interpretability. The suggested framework has a high potential to be a clinical decision-support system among dermatologists, which will help detect skin diseases earlier and more accurately.