Quantum Feature Optimization for Enhanced Clustering of Blockchain Transaction Data

Yun-Cheng Tsai, Samuel Yen-Chi Chen

Abstract

Blockchain transaction data presents significant challenges for classical clustering due to its high dimensionality and non-linear feature relationships. This paper explores the application of Quantum Machine Learning (QML) to enhance the clustering of financial blockchain analytics. By leveraging Quantum Neural Networks (QNNs) to map data into high-dimensional Hilbert spaces, we demonstrate substantial improvements in cluster separation. Our results show that even shallow, untrained quantum circuits can significantly outperform classical K-Means baselines, with optimized quantum feature spaces achieving near-perfect Silhouette Scores.

Introduction

Analyzing blockchain transactions is critical for identifying fraudulent activity and modeling economic behavior. However, the "entangled" nature of features in these datasets often renders traditional Euclidean-based clustering algorithms, such as K-Means, ineffective. Classical methods struggle to capture the complex, non-linear structures inherent in financial data. This study investigates the efficacy of quantum-enhanced clustering strategies, comparing classical, hybrid, and fully quantum approaches to provide a systematic evaluation of QML in blockchain analytics.

Methodology

The research framework evaluates three distinct clustering strategies to identify the most effective representation for transaction data:

  1. Traditional K-Means: Utilizing standard pre-processed classical features as a baseline.
  2. Hybrid Quantum-Classical: Concatenating classical features with "quantum random features" generated by an untrained, randomly initialized QNN.
  3. Fully Quantum Clustering: Employing a QNN trained with a SwAV (Swapping Assignments between Views) self-supervised loss function to optimize the quantum feature space for maximum cluster distinctness.
Methodological workflow comparing classical, hybrid, and fully quantum clustering strategies.

The quantum components utilize variational circuits composed of single-qubit $R_y$ rotations and CNOT gates for entanglement. This architecture projects data into a higher-dimensional Hilbert space, where latent structures that are "tangled" in classical space become linearly separable.

Results and Discussion

The integration of quantum features led to a dramatic increase in clustering performance across all metrics. While classical features yielded a Silhouette Score of 0.265 for $K=2$, the fully optimized QNN approach reached a near-perfect 0.999. Furthermore, the compactness of the clusters improved significantly; the Davies-Bouldin Index dropped from 1.525 (classical) to 1.11e-8 (quantum).

A key insight from the study is the effectiveness of quantum depth. As illustrated in Figure 3, even a quantum depth of 1 provides a massive performance boost over the classical baseline (represented at depth 0). This suggests that the inherent non-linearity of quantum transformations is a powerful tool for feature extraction, even before optimization or in the presence of NISQ-era hardware constraints.

Average Silhouette Score across different quantum depths and cluster counts (K), showing the immediate benefit of quantum feature extraction over the classical baseline (depth 0).

The consistency of these results across various cluster counts ($K=2$ to $K=6$) underscores the robustness of quantum-enhanced methods. The self-supervised SwAV approach further demonstrates that QNNs can learn high-quality representations without the need for labeled ground truth data, which is often unavailable in blockchain contexts.

Conclusion

Quantum feature optimization offers a transformative approach to blockchain data analysis. By mapping transactions into quantum Hilbert spaces, we achieve superior cluster definition and separation compared to traditional classical methods. While computational overhead and scalability remain challenges for current hardware, the significant performance gains—even with shallow circuits—point toward a promising future for quantum-enhanced financial monitoring and fraud detection.

Summary updated

BibTeX
@inproceedings{tsai2025quantum,
  author = {Yun-Cheng Tsai and Samuel Yen-Chi Chen},
  title = {{Quantum Feature Optimization for Enhanced Clustering of Blockchain Transaction Data}},
  booktitle = {ICCCN},
  year = {2025},
  doi = {10.1109/icccn65249.2025.11133860},
}
Figure