The classification of museum artworks by medium is a crucial yet challenging task, often hindered by class imbalance and the variability of visual patterns in datasets. The purpose is to improve the accuracy and robustness of artwork classification through a ensemble learning framework. The proposed method combines the ResNet101, DenseNet121, and VGG19 architectures, using their complementary strengths in feature extraction and generalization. Using the MAMe (Museum Art Medium) dataset, which comprises 29 medium classes and high-resolution images, the study implements comprehensive preprocessing techniques, including data augmentation and resizing input data, to optimize computational efficiency while preserving critical visual features. The experimental results show that ensemble model outperformed individual models, with a training accuracy of 84.45% and a testing accuracy of 81.68%. This approach provides a scalable and domain-specific solution for artwork classification. Future research will explore the integration of multimodal data, such as textual and historical metadata, to further enhance the classification and analysis capabilities.