Quantum SVM cuts false-positive rate vs. random forest but slightly lowers true-positive rate
The claim
A quantum support vector machine reduces FPR while marginally decreasing TPR compared to a random forest in QML intrusion detection
Published as: “False-Alarm Improvements and Detection Trade-Offs in QML Intrusion Detection: A Secondary Analysis of the Kaissar Corpus”
So what
The analysis shows that reported low false-positive rates in QML-IDS papers often lack independent-test validation and cannot be directly translated into operational guarantees.
The five-minute version
A quantum support vector machine (QSVM) reduces the false-positive rate compared with a random forest (3.329% vs 6.875%) but its true-positive rate drops slightly (93.682% vs 93.955%). The secondary analysis reconstructs joint FPR/TPR operating points from five QML-IDS papers, finding that QSVM also lowers false alarms against other classical baselines, though sometimes at the cost of detection (e.g., against random forest). Evidence is limited to 24 descriptive points with conditional confidence intervals; many points lack verified independent-test independence and sample sizes are too small to support ultra-low FPR claims such as 0.01%.
What they found 3
- QSVM reduces FPR compared with random forest but also slightly lowers TPR
“In one test-labelled comparison, a quantum support vector machine reduces FPR from 6.875% for random forest to 3.329%, while TPR decreases from 93.955% to 93.682%; both rates improve against three other comparators.”
- QSVM-IGWO has fewer false positives but more false negatives than random forest
“The QSVM with improved grey wolf optimization (QSVM-IGWO) point has 308 false positives and 716 false negatives; the random forest (RF) point has 636 and 685.”
- K10’s benign row yields an observed FPR of 0.200% with a conditional upper 95% limit of 0.324%
“This identifies FPR 0.200% and a conditional upper limit of 0.324%.”
What undercuts it the paper’s own results
- The QSVM’s lower false-alarm rate accompanies one additional missed attack, rather than an improvement in every measure.
- Their sample sizes cannot support the 0.01% reference with the specified upper limit even in a zero-error case.
Stated limits 3
- Other eligible count points remain conditional illustrations with unverified independence: these are K10 and K17, not a validated independent-test tier.
- Their sample sizes cannot support the 0.01% reference with the specified upper limit even in a zero-error case.
- All rates are retained, while the same provenance rule places K03 and K11 intervals in a hypothetical supplement.
What’s new here 4
- Reconstructs joint false-positive and true-positive rates from published QML-IDS studies rather than reporting only aggregate metrics.
- Applies a provenance rule to separate descriptive, training-only, and conditional (independence-unverified) results.
- Provides conditional Clopper-Pearson confidence intervals for reconstructed rates and evaluates their ability to meet analytical low-FPR references.
- Highlights trade-offs where reduced false alarms accompany decreased detection, clarifying the operational relevance of reported improvements.
Artifacts & methods 10
- quantum support vector machine (QSVM)system
- QSVM with improved grey wolf optimization (QSVM-IGWO)system
- random forest (RF)system
- logistic regression (LR)system
- k-nearest neighbours (KNN)system
- decision tree (DT)system
- support vector machine (SVM)system
- convolutional neural network (CNN)system
- quantum convolutional neural network (QCNN)system
- BoT-IoT datasetdataset