Research on Abnormal Radio Detection Method Combining Local Outlier Factor and One-Class Support Vector Machine

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

1. the weight and threshold settings should be more "transparent"
Now the fusion weight wlof, wocsvm, adaptive threshold (the critical quantile obtained by KDE), low confidence interval parameter δ and model difference threshold Δ difference are all described, but "how to select" and "sensitivity to the results" are missing.
Suggestions for modification: give the specific selection process of weight/threshold (for example, maximize F1 or AUC with 50% CV on the training set)
2. baseline fairness: input and adjustment parameters should be consistent as much as possible
In this paper, the depth baseline (deep SVDD, e-gan) directly eats the original spectrum, while the traditional methods (if, Aine, lof-ocsvm) use 12 dimensional features, which is easy to "feed the number is inconsistent".
Suggestions for modification: it is suggested to supplement a version of "comparison with input": either give the traditional method a version of "original spectrum+end-to-end" (even simple CNN), or add the version of "feature → shallow model" to the depth baseline, or at least discuss the "influence of input mode differences on results" separately in the method section.
3. complete simulation of data set, supplemented by a group of real (or semi real) verification
At present, the data is from the simulated OFDM scene and the interference is superimposed on the designated ISR; This is a good starting point, but it is easy to "over fit the simulation distribution".
Suggestions for modification: (a) repeat the core indicators with a small segment of real spectrum (or public data); (b) Perform "extraterritorial" generalization test (such as changing carrier spacing/frequency hopping strategy) and report the amplitude loss of f1/auc. This can significantly enhance persuasion.
4. replenishment of reproducibility and report granularity
Suggestions for modification: list the key superparameters (LOF neighborhood number k, OCSVM kernel/γ/ν, KDE bandwidth, etc.) and random seeds.
5. consistency of charts and terms
(a) "Lognormal/log normal/lognormal" appears at the same time, and is unified as "log normal";
(b) The "9 features/10 (FE)/10 (fr-mi)/10 (log normal)/12 features" in Table 1 suggests that the specific dimensions and new items of each should be clearly defined in the notes to the table, which readers can understand at a glance.
6. the summary converges to "one proposition+a set of key figures"
The current summary uses the entry sentence pattern of "feature effectiveness/model superiority/robustness:" to slightly show the tone of "publicity".
Suggestions for revision: revised to twoorthree sentences fluent Chinese/English: one sentence about problems and methods, one sentence about core numbers (such as − 20 dB recall/F1/AUC comparison), and one sentence about generalization/robustness.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Please see the comments in the attached document.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This paper proposes a novel radio anomaly detection method that integrates fluctuation entropy, fluctuation mutual information, and log-normal distribution fitting parameters with a fusion model combining LOF and OCSVM. Simulation results demonstrate that the proposed approach achieves superior performance under low interference-to-signal ratio conditions, significantly improving recall and F1-score compared to baseline features and outperforming existing models, such as Deep SVDD and E-GAN. However, several concerns and issues should be addressed. 

1. Provide a detailed comparison of computational cost with baseline methods, as real-time spectrum monitoring often requires lightweight solutions.
2. Conduct experiments on key parameter sensitivities (e.g., LOF neighbor size, OCSVM ν, δ thresholds).
3. Discuss practical implementation issues, such as computational optimization and suitability for embedded or hardware-constrained spectrum monitoring systems.
4. Incorporate real-world or publicly available datasets to demonstrate robustness and generalizability beyond simulation.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

All my previous comments have been resolved.