Research Title:
Chronological Evaluation of a Hybrid Quantum–Classical Graph Neural Network for Illicit Bitcoin Transaction Detection
Research Problem:
Financial transaction networks are naturally represented as graphs because transactions are connected through transaction flows. Graph Neural Networks can exploit these relationships for illicit transaction detection. At the same time, quantum machine learning has been proposed as a potential extension to classical machine learning. However, quantum components should not be assumed to provide an advantage without direct experimental comparison. This research therefore evaluates a classical GCN against a hybrid GCN–VQC model under a chronological Bitcoin transaction setting.
Dataset:
Elliptic Bitcoin Transaction Dataset
Dataset Characteristics:
• Total transactions/nodes: 203,769
• Directed transaction-flow edges: 234,355
• Model input features per transaction: 165
• Total time steps: 49
• Classes used for supervised learning: Licit and Illicit
• Unknown-labelled transactions: Excluded from supervised learning/evaluation
Input to the Model:
The input consists of 165-dimensional feature vectors for Bitcoin transaction nodes together with the graph connectivity representing transaction flow between transactions. Therefore, the GCN learns from both individual transaction characteristics and relational information from connected transactions.
Preprocessing:
• Transaction ID and time information were excluded from model input features.
• Unknown-labelled transactions were excluded from the supervised classification task.
• StandardScaler was fitted using training-period data only.
• The fitted scaler was subsequently applied to validation and test data.
• Separate graph objects were constructed for the chronological periods.
Chronological Data Split:
Training:
• Time steps: 1–34
• Nodes: 29,894
• Edges: 22,898
Validation:
• Time steps: 35–41
• Nodes: 7,829
• Edges: 6,509
Held-out Test:
• Time steps: 42–49
• Nodes: 8,841
• Edges: 7,217
Training Class Distribution:
• Licit transactions: 26,432
• Illicit transactions: 3,462
• Class weighting was used because of the class imbalance.
Classical Model:
A two-layer Graph Convolutional Network was implemented as the classical baseline.
Architecture:
165 input features
→ GCN layer (165 → 32)
→ ReLU
→ GCN layer (32 → 2)
→ Licit/Illicit prediction
Trainable Parameters:
5,378
Hybrid Quantum–Classical Model:
The hybrid model combines graph-based feature extraction with a Variational Quantum Circuit.
Architecture:
165 input features
→ GCN layer (165 → 32)
→ ReLU
→ graph representation
→ linear projection (32 → 4)
→ 4-qubit Variational Quantum Circuit
→ 2 variational layers
→ quantum expectation-value measurements
→ classical output classifier
→ Licit/Illicit prediction
Quantum Configuration:
• Number of qubits: 4
• Variational layers: 2
• VQC trainable parameters: 8
• Hybrid model total trainable parameters: 6,518
• Quantum execution: simulator
• Random seed: 42
• Differentiable quantum training: 256 selected training nodes because of CPU simulation cost
Evaluation Metrics:
Because the dataset is highly imbalanced, accuracy alone was not used as the main indicator. The models were evaluated using:
• AUPRC
• ROC-AUC
• F1-score
• Precision
• Recall
Held-Out Test Results:
Classical GCN:
• AUPRC: 0.145466
• ROC-AUC: 0.771520
• F1-score: 0.146276
• Precision: 0.080482
• Recall: 0.801471
Hybrid Quantum Model:
• AUPRC: 0.038754
• ROC-AUC: 0.448034
• F1-score: 0.066192
• Precision: 0.035916
• Recall: 0.421569
Main Findings:
1. The classical GCN outperformed the tested hybrid quantum model on every reported held-out test metric.
2. The classical GCN achieved substantially higher AUPRC and ROC-AUC than the hybrid model.
3. Classical recall reached 0.801471, showing that it identified a relatively large proportion of illicit transactions at the fixed 0.5 decision threshold.
4. Classical precision remained low at 0.080482, indicating that false positives remain an important problem.
5. The tested 4-qubit VQC did not improve predictive performance over the classical GCN.
6. No quantum performance advantage was demonstrated in this experimental configuration.
7. Quantum simulation introduced a practical computational limitation, requiring the differentiable quantum stage to be restricted to 256 training nodes.
8. Chronological splitting and training-only feature scaling reduced obvious future-information leakage compared with a random-split approach.
What the Research Establishes:
The experiment establishes that, under the implemented chronological Elliptic Bitcoin transaction setting, the classical GCN provided better held-out predictive performance than the evaluated 4-qubit hybrid GCN–VQC configuration. The result provides empirical evidence that adding a quantum component does not automatically improve a graph-based financial transaction classifier.
The research does NOT establish that all classical GNNs are superior to all quantum GNNs. The conclusion applies only to the architecture, dataset, training conditions, and quantum configuration evaluated in this experiment.
Limitations:
• Only one random seed (42) was evaluated.
• Classical and hybrid models were not parameter matched (5,378 vs 6,518 parameters).
• Differentiable quantum training was restricted to 256 selected training nodes.
• Quantum execution used CPU simulation rather than real quantum hardware.
• Only one 4-qubit, 2-layer VQC architecture was evaluated.
• The experiment used one dataset.
• A fixed classification threshold of 0.5 was used.
• Archived quantum validation records contained an inconsistency; therefore, disputed quantum validation values were excluded from the main quantitative claims.
• Stronger temporal GNN architectures such as EvolveGCN or TGN were not compared under identical preprocessing.
• Federated Learning and FedAvg were NOT implemented in this experiment; therefore, no federated-learning, privacy-preserving, or cross-institutional performance claim is made.
Final Conclusion:
Under chronological and class-imbalance-aware evaluation on the Elliptic Bitcoin transaction graph, the classical GCN performed better than the tested hybrid quantum–classical model across AUPRC, ROC-AUC, F1-score, precision, and recall. The evaluated quantum component did not provide a performance advantage in this setting. These findings demonstrate the importance of experimentally validating proposed quantum advantages rather than assuming that the addition of a quantum circuit will necessarily improve financial fraud or illicit-transaction detection.