隨著深度學習技術的快速進展，異常偵測在工業製程等高複雜度資料情境中，面臨資料結構階層性強與類別分布極度不平衡的雙重挑戰。此類問題常見於跨階層依賴關係明顯且資料異質性高的應用場域，如感測器資料、製程歷程紀錄與零件維修履歷等。傳統模型在處理跨層級相依與異質特徵交互方面常受限，無法有效辨識隱藏於多階層關聯中的異常樣本，進而影響異常辨識的靈敏度與整體穩健性。
為解決上述問題，本研究提出一套具階層感知能力之圖神經網路 (Graph Neural Network, GNN) 異常偵測架構，稱為階層式圖異常偵測模型 (Hierarchical Graph-based Anomaly Detection, HiGAD)，能有效建模多層級結構中的節點間相依關係與非線性交互特性。模型核心採用圖注意力機制 (Graph Attention Network, GAT) ，透過層次化圖結構設計與逐層特徵聚合策略，使網路能動態掌握跨階層特徵傳遞與上下層節點間的影響關係，強化模型在階層式資料中的特徵學習能力與語意整合表現。此外，本研究引入條件變分自動編碼器 (Conditional Variational Autoencoder, CVAE) ，以擴展結構一致性之潛在特徵表示，藉由正規化潛在空間與樣本分佈，使得異常樣本在嵌入空間中的表現更具可辨識性。針對資料不平衡問題，本研究進一步整合 K-means SMOTE 技術於潛在空間中進行過採樣操作，依據局部群內樣本密度進行樣本生成，有效提升少數類別異常樣本的資料擴展品質與學習穩定性。
實驗部分，本研究選用多組具階層結構與高度不平衡分佈之工業製程資料集進行驗證，並採用混淆矩陣、分類評估指標與 AUC 曲線等進行模型效能評估。結果顯示，本研究所提出之 HiGAD，無論在異常偵測的 Precision、Recall或整體穩定性方面，皆顯著優於傳統深度學習方法與不具階層建模能力之 GNN 架構。特別是在樣本極度不平衡或多層級特徵交互複雜度高的資料情境中，本模型展現出更高的辨識靈敏度與泛化能力。
綜上所述，本研究展示 HiGAD 於階層結構資料異常偵測任務中的潛在優勢，並透過潛在特徵擴展與結構化過採樣策略，有效提升模型在實務應用中對異常樣本的辨識準確性與穩健性。所提方法兼具理論創新與應用價值，未來可進一步延伸應用至智慧製造、設備預測維護與高風險工序監控等具結構依賴特性的領域。

The advancement of deep learning has brought new opportunities to anomaly detection in complex industrial data, yet significant challenges remain due to hierarchical data structures and severe class imbalance. These issues are common in scenarios with cross-level dependencies and high data heterogeneity, such as sensor readings, process logs, and maintenance records. Conventional models often struggle to capture multi-level correlations and non-linear feature interactions, limiting both sensitivity and robustness.
To tackle these challenges, we propose HiGAD (Hierarchical Graph-based Anomaly Detection), a novel graph neural network framework designed with hierarchical awareness. Leveraging Graph Attention Network (GAT), HiGAD effectively models inter-layer dependencies through a structured graph design and progressive feature aggregation, enabling more accurate cross-level semantic learning. A Conditional Variational Autoencoder (CVAE) is further integrated to enhance latent representation consistency and improve anomaly separability within the embedding space.
To address class imbalance, we adopt K-means SMOTE in the latent space to synthesize minority-class samples based on local sample density, thereby improving data diversity and training stability. Experiments conducted on multiple real-world industrial datasets demonstrate that HiGAD consistently outperforms conventional deep learning and non-hierarchical GNN models in precision, recall, and overall generalization, particularly under highly imbalanced and structurally complex conditions.

Aggarwal, C. C., & Sathe, S. (2017). Theory of outlier ensembles. Outlier Ensembles: An Introduction, 35-74.
An, J., & Cho, S. (2015). SNU data mining center 2015-2 special lecture on IE variational autoencoder based anomaly detection using reconstruction probability, Soeul National University. The Journal of Machine Learning Research, 3-4.
Araf, I., Idri, A., & Chairi, I. (2024). Cost-sensitive learning for imbalanced medical data: A review. Artificial Intelligence Review, 57, 80.
Bishop, C. M. (2006). Pattern Recognition and Machine Learning. Springer.
Chandola, V., Banerjee, A., & Kumar, V. (2009). Anomaly Detection: A Survey. ACM Computing Surveys, 41(3), 1-58.
Chawla, N. V., Bowyer, K. W., Hall, L. O., & Kegelmeyer, W. P. (2002). SMOTE: Synthetic Minority Over-sampling Technique. Journal of Artificial Intelligence Research, 16(16), 321–357.
Douzas, G., Bacao, F., & Last, F. (2018). Improving imbalanced learning through a heuristic oversampling method based on k-means and SMOTE. Information Sciences, 465, 1–20.
Ekle, O. A., & Eberle, W. (2024). Anomaly detection in dynamic graphs: A comprehensive survey. ACM Transactions on Knowledge Discovery from Data, 18(8), 1-44.
Gim, U., & Park, Y. (2022). Latent Vector Expansion using Autoencoder for Anomaly Detection. In Proceedings of the 34th Workshop on Image Processing and Image Understanding (IPIU 2022).
Goodfellow, I. J., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., & Bengio, Y. (2014). Generative Adversarial Nets. Advances in Neural Information Processing Systems, 27.
Hua, S., Li, X., Jing, Y., & Liu, Q. (2022). A semantic hierarchical graph neural network for text classification. arXiv preprint. ArXiv:2209.07031.
Kingma, D. P., & Welling, M. (2014). Auto-Encoding Variational Bayes. In International Conference on Learning Representations (ICLR 2014).
Kipf, T. N., & Welling, M. (2017). Semi-Supervised Classification with Graph Convolutional Networks. In Proceedings of the 5th International Conference on Learning Representations (ICLR 2017).
Liu, F. T., Ting, K. M., & Zhou, Z. H. (2008). Isolation forest. In 2008 eighth ieee international conference on data mining, 413-422.
Liu, X.-Y., Wu, J., & Zhou, Z.-H. (2009). Exploratory Undersampling for Class-Imbalance Learning. IEEE Transactions on Systems, Man, and Cybernetics, Part B (Cybernetics), 39(2), 539–550.
Qiao, H., Tong, H., An, B., King, I., Aggarwal, C., & Pang, G. (2024). Deep Graph Anomaly Detection: A Survey and New Perspectives . arXiv preprint. ArXiv:2409.09957.
Sakho, A., Malherbe, E., Gauthier, C.-E., & Scornet, E. (2025). Harnessing Mixed Features for Imbalance Data Oversampling: Application to Bank Customers Scoring . arXiv preprint. ArXiv:2503.22730.
Shi, Y., Huang, Z., Feng, S., Zhong, H., Wang, W., & Sun, Y. (2021). Masked Label Prediction: Unified Message Passing Model for Semi-Supervised Classification. International Joint Conference on Artificial Intelligence (IJCAI 2021), 1548-1554.
Sohn, K., Lee, H., & Yan, X. (2015). Learning structured output representation using deep conditional generative models. Neural Information Processing Systems, 28.
Tang, J., Li, J., Gao, Z., & Li, J. (2022). Rethinking Graph Neural Networks for Anomaly Detection. Proceedings of the 39th International Conference on Machine Learning (ICML 2022), 21076–21089.
Veličković, P., Cucurull, G., Casanova, A., Romero, A., Liò, P., & Bengio, Y. (2018, February 4). Graph Attention Networks. Proceedings of the 6th International Conference on Learning Representations (ICLR 2018).
Wang, A. X., Le, M. Q., Duong, H.-T., Van, B. N., & Nguyen, B. P. (2025). CTVAE: Contrastive Tabular Variational Autoencoder for imbalance data. Knowledge and Information Systems, 1-20.
Wang, S., & Yu, P. S. (2022). Graph neural networks in anomaly detection. Graph neural networks: Foundations, frontiers, and applications, 557-578.
Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., & Yu, P. S. (2020). A Comprehensive Survey on Graph Neural Networks. IEEE Transactions on Neural Networks and Learning Systems, 32(1), 4–24.
Zhong, Z., Li, C.-T., & Pang, J. (2023). Hierarchical message-passing graph neural networks: Data Mining and Knowledge Discovery. Data Mining and Knowledge Discovery, 37(1), 381–408.
Zhu, R., Guo, Y., & Xue, J.-H. (2020). Adjusting the imbalance ratio by the dimensionality of imbalanced data. Pattern Recognition Letters, 133, 217–223.
Zong, B., Song, Q., Renqiang, M., Cheng, W., Lumezanu, C., Cho, D., & Chen, H. (2018). Deep Autoencoding Gaussian Mixture Model for Unsupervised Anomaly Detection. Proceedings of the 6th International Conference on Learning Representations (ICLR 2018).