| Title |
Ensuring Blockchain Validator Reliability through Multivariate Time-series Anomaly Detection |
| Authors |
이승재(Seungjay Lee) ; 김명선(Myungsun Kim) |
| DOI |
https://doi.org/10.5573/ieie.2026.63.2.35 |
| Keywords |
Anomaly detection; Dimension reduction; Blockchain validator |
| Abstract |
Blockchain validators are crucial for ensuring the availability and integrity of blockchain networks. However, research on maintaining validator reliability has not yet been active. This paper proposes a deep learning-based anomaly detection framework specifically designed for blockchain validators, addressing challenges arising from their high-dimensional telemetry data. To overcome the ‘curse of dimensionality,’ dimensionality reduction techniques are applied as a preprocessing step for anomaly detection. We systematically evaluate combinations of various dimension reduction techniques (PCA, UMAP, LLE, ISOMAP) and anomaly detection models (LSTM-VAE, VTTSAT, VTTPAT). Experimental results show that the combination of PCA and VTTSAT achieved the best performance with a PA%K AUC of 0.7922. Furthermore, experiments varying the reduced dimension count revealed that 50 dimensions provided the most effective representation for validator anomaly detection. Our findings highlight the feasibility of applying deep learning-based anomaly detection in validator environments, demonstrating its potential to enhance the reliability of blockchain operations. |