Coffee aroma is characterized by complex and strongly overlapping mixtures of volatile organic compounds (VOCs), which poses substantial challenges for reliable classification. In this study, a temperature-modulated metal oxide sensor array with four sensing channels is used to classify the aromas of commercially available coffee capsule products. A feature-based Random Forest classifier was employed as a conventional baseline and compared with a regularized one-dimensional convolutional neural network (1D CNN) that operates directly on raw sensor time series. While the Random Forest model achieves satisfactory classification performance, the CNN consistently yields higher accuracy by learning discriminative temporal response patterns without relying on handcrafted features, achieving a maximum classification accuracy of 99.82%. Owing to the short measurement duration and compact sensing configuration, the proposed approach enables rapid classification of coffee aromas. The results demonstrate that compact convolutional neural networks, combined with four-channel MOX sensor arrays, provide an effective and scalable approach for the classification of commercial coffee capsule products. This work highlights the potential of low-cost electronic nose systems for practical product-level aroma discrimination under controlled measurement conditions.