A new research paper explores the effectiveness of Quantum Kernel Support Vector Machines (QSVM) for credit card fraud detection, a task complicated by the rarity of fraudulent transactions. The study benchmarks three quantum encoding methods against classical models, embedding synthetic minority oversampling (SMOTE) into the quantum workflow to address class imbalance. While classical models achieved a PR-AUC of 0.983 on a large dataset, their performance dropped significantly to 0.750 when constrained to match QSVM hardware limitations. The best quantum result achieved a PR-AUC of 0.730, with further performance degradation to 0.62 on physical quantum hardware due to noise and gate errors. The research identifies the scaling of Gram matrix evaluation and feature compression onto limited qubit registers as key bottlenecks for near-term quantum kernels. AI
IMPACT Highlights limitations of current quantum hardware for practical ML tasks, suggesting classical methods remain superior for complex problems like fraud detection.
RANK_REASON Research paper published on arXiv detailing experimental results. [lever_c_demoted from research: ic=1 ai=1.0]
- Angle
- Credit Card Fraud Detection: An Evaluation of SMOTE Resampling and Machine Learning Model Performance
- Europe
- Hilbert spaces
- Quantum Kernel Support Vector Machines
- Quantum Kernel SVMs
- Quantum Machine Learning
- SMOTE
- ZZFeatureMap
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
