Quantum circuits reduce data imbalance recall gap by 10-25%

A new hybrid quantum-classical framework has demonstrated visible improvements in machine learning performance on unbalanced datasets, with researchers reporting 5-15% increases in F1 scores and 10-25% improvements in minority class recall. A team consisting of Tanapol Nuatho, Narisorn Sangnakara, Prapong Prechaprapranwong, and Rajchawit Sarochawikasit has developed a quantum circuit bone machine (QCBM) to generate synthetic data that addresses limitations caused by data scarcity and class imbalance. Experiments utilizing the Iris and Telco Customer Churn datasets show that enriching the training data with QCBM-generated samples with 40-50% of the minority classes improves the F1 score by approximately 5-15%. Cross-domain evaluation revealed a performance gap of only 3–10%, indicating high distribution fidelity of the quantum-generated samples.

QCBM for unbalanced tabular data generation

A 5-15 percent increase in F1 score and a 10-25 percent improvement in minority class recall are now achievable by enriching training data with synthetic samples generated using new applications of quantum computing, specifically quantum circuit bone machines (QCBMs). Researchers are tackling the long-standing challenge of class imbalance in machine learning. In this challenge, algorithms struggle to accurately identify rare but important instances, a common problem in areas such as fraud detection and medical diagnostics. This performance improvement indicates a measurable benefit directly related to the utilization of quantum mechanical principles. The team’s approach focuses on QCBMs, which are parameterized quantum circuits designed to model complex probability distributions. Unlike classical generative methods, QCBM exploits quantum superposition and entanglement to potentially capture subtle relationships in the data.

Before quantum processing, the data undergoes classical preprocessing. Regularization and principal component analysis (PCA)-based dimensionality reduction are employed to enable efficient basis encoding of quantum circuits. This hybrid quantum-classical pipeline is important for converting real-world tabular data into a format suitable for quantum computation. The QCBM is then trained by minimizing the Kullback-Leibler (KL) divergence, which is a measure of the difference between the actual and generated data distributions, using gradient-based parameter shift optimization rules. This is an important finding because the fidelity of the data generated is surprisingly high and the usefulness of synthetic data depends on accurately representing the underlying distribution.

PCA-based dimensionality reduction for quantum encoding

Efficient data encoding is essential for preparing classical data for processing on quantum hardware, and this task is essential to realizing the potential of quantum machine learning. Quantum Circuit Born Machine (QCBM) offers a new approach to synthetic data generation, but its effectiveness depends on this efficient encoding. This dimensionality reduction is not an arbitrary simplification of the data. It means fitting classical information to the constraints of quantum representation. By reducing the number of features, PCA minimizes the number of qubits needed to encode the data. This is an important consideration given the limited availability and inherent vulnerabilities of qubits. This careful preparation allows for a more efficient mapping of classical data to the quantum domain, maximizing the potential of QCBM. This preprocessing step does not sacrifice performance, and the improved recall is especially important for datasets suffering from class imbalance, where identifying instances of minority classes is paramount.

The researchers also found that QCBM achieved competitive classification performance and produced lower maximum mean discrepancy (MMD) on the news agency dataset. This shows superior structural similarity compared to traditional oversampling techniques. This combination of improved recall, distributional fidelity, and structural similarity establishes QCBM as a promising tool to address data scarcity and imbalance in machine learning applications when combined with careful conventional preprocessing.

Distribution fidelity with KL divergence and MMD

Researchers have refined a method to assess the quality of synthetic tabular data produced by quantum circuit bone machines (QCBMs), moving beyond simple classification accuracy to focus on distributional fidelity. While achieving high classification scores on augmented datasets is important, the team emphasizes the need to verify how closely the synthetic data reflects the statistical properties of the original real-world data. This is especially important when dealing with the challenge of unbalanced datasets, where minority classes are underrepresented and require careful expansion to avoid biased models. In parallel to KL divergence, the team utilizes maximum mean discrepancy (MMD) to assess structural similarity, providing a complementary perspective on the fidelity of the generated data. The researchers systematically compare QCBM with four SMOTE variants and say they provide the first comprehensive classical quantum benchmark for tabular data augmentation.

Results show that QCBM not only achieves competitive classification performance but also produces consistently low MMD scores on the carrier customer churn dataset, suggesting a good ability to capture the underlying data structure. The evaluation extends to cross-domain testing to evaluate the generalization ability of QCBM. Through “train on synthetic, test on real” (TSTR) and “train on real, test on synthetic” (TRTS) protocols, the team observed performance gaps of just 3-10% and demonstrated an incredibly high level of distributional fidelity. This suggests that the synthetic data generated by QCBM is not only useful for improving classification, but also that its statistical properties closely resemble real-world data. The researchers report that enriching the training data with QCBM-generated synthetic samples representing 40% to 50% of the minority classes improves recall for the minority classes by 10% to 25%, a significant benefit for applications where accurate identification of rare events is critical.

Improve F1 score and recall with synthetic samples

Using quantum computing techniques to enrich unbalanced datasets with synthetically generated samples can measurably improve machine learning performance, especially in scenarios where minority class identification is important. Researchers have demonstrated that applying quantum circuit bone machines (QCBM) to datasets struggling with non-uniform class representation significantly improves recall by 10-25% and F1-score by 5-15%. This advantage is particularly valuable in areas such as fraud detection and medical diagnostics, where missing even a small number of positive cases can have serious consequences. QCBM, a parameterized quantum circuit, does more than simply duplicate existing data points. Learn the underlying probability distribution and generate new samples that closely reflect real data. This fidelity is proven by cross-domain evaluations, where models trained on QCBM-generated data perform with only a 3,10% performance gap when tested on real-world datasets, and vice versa.

This suggests that the synthetic data does not simply mimic the training set, but captures essential characteristics of the broader data distribution. The ability to generate high-fidelity synthetic data opens new avenues to improve model generalization and reduce bias in critical applications.

Cross-domain evaluation of QCBM performance

A subtle application of quantum computing, namely synthetic data generation, is gaining attention. This cross-domain performance, where the model is trained on QCBM-generated data and evaluated on real-world examples, challenges the assumption that quantum-generated data is inherently limited to the characteristics of its training source. The researchers focused on evaluating QCBM’s ability to generalize beyond the dataset used for initial training. Specifically, the researchers reported an increase in F1 score of approximately 5% to 15% and a 10% to 25% improvement in recall for minority classes. This is an important metric when dealing with unbalanced datasets where identifying rare events is paramount. This suggests that QCBM is not simply remembering training data, but learning underlying patterns that can be applied to new, unknown instances. The team went beyond evaluating performance within the same dataset.

Cross-domain evaluation reveals a performance gap of only 3,10%, indicating strong distributional fidelity. This level of fidelity is particularly important because it suggests that QCBM is able to capture the essential features of the data, rather than producing statistically valid but structurally distinct samples. Comparative analysis further strengthened QCBM’s position. These findings establish QCBM as a viable complementation tool for data augmentation, especially for low-dimensional structured table data with class imbalance.

Researchers carefully evaluated QCBM against four well-established classical oversampling techniques: SMOTE, Borderline-SMOTE, KMeans-SMOTE, and SVM-SMOTE in a comprehensive benchmark designed to assess the viability of quantum approaches for tabular data augmentation. This improvement is particularly important given the inherent difficulty in accurately predicting rare events or identifying minority groups.