Quantum Feature Engineering for Credit Default Prediction: When and Why IQP Circuits Help Linear Classifiers
Exploring practical bounds for quantum machine learning on tabular data, researchers tested Instantaneous Quantum Polynomial-time (IQP) circuits as feature extractors for credit default prediction. On the UCI credit card dataset, appending 16 IQP features from an 8-qubit circuit lifted Logistic Regression's F1 score from 0.462 to 0.517, outperforming Kernel PCA at an identical feature budget. Crucially, non-linear models like XGBoost and Random Forest saw zero improvement. The experiment demonstrates that the quantum circuit does not fabricate signal, but instead provides an explicit non-linear projection that compensates specifically for the architectural limits of linear classifiers.
Why it's worth reading
By precisely documenting where IQP circuits help and where they fail to add value over tree models, the paper provides a grounded, hype-free benchmark for quantum feature engineering on tabular data.