<p>Seismic facies interpretation in the Arctic continental margin is essential for understanding complex subsurface structures shaped by permafrost, gas hydrates, and glacial-interglacial climatic fluctuations. Conventional seismic facies interpretation requires substantial effort and is highly susceptible to interpreter subjectivity. To overcome these limitations, this study proposes an unsupervised learning framework that integrates multi-attribute analysis with a convolutional autoencoder for automatic seismic facies classification along the western Chukchi Rise, Arctic Ocean. The workflow consists of three components: (1) extraction of seismic attributes, (2) feature learning using an autoencoder, and (3) dimensionality reduction through Uniform Manifold Approximation and Projection (UMAP). The reduced embeddings were subsequently clustered using hierarchical agglomerative clustering (HAC), Gaussian mixture model (GMM), and self-organizing map (SOM), and their quantitative performance was evaluated. Through the numerical experiments, the autoencoder effectively captures latent structural and textural patterns, while UMAP preserves meaningful relationships among attributes during dimensionality reduction. Among the three clustering methods, HAC achieved the highest performance, successfully distinguishing seismic facies based on subtle variations in amplitude and lateral continuity. The proposed approach improves the objectivity and efficiency of seismic facies classification for characterizing Arctic subsurface structures. These findings demonstrate that the proposed framework can reliably delineate seismic facies within the geologically complex Arctic environment.</p>