Anomaly Detection in the Internet of Vehicular Networks Using Explainable Neural Networks (xNN)

Round 1

Reviewer 1 Report

Starting with 2011, all attempts to analyze data and correct cyber intrusions and attacks were briefly presented, and the proposed method is convincingly presented with optimal results.

It is recommended that in the mathematical formulas of the optimal model chosen xNN, the following aspects be specified more clearly:

- the role of σ (Sigma) - is the global constant, discovered by experimental analysis with xNN simulation? What percentage of influence does it have in the elaborate mathematical system?

- the correction factors introduced by h each have a clear notation and an explanation (example for "disturbance coverage", etc.)

It is also recommended that a conditional block of type «IF» be introduced in the logic scheme of the optimal model chosen xNN, because the mathematical iterations are executed and repeated until certain optimal values are reached.

The authors could also conclude that the data package in the protocol analyzed with maximum values by applying the high-performance xNN model would be preferable to be used more widely in the future, to reduce and eliminate security attacks such as IoV - generated by the hacker.

Author Response

Please see the attachment, thanks.

Author Response File: Author Response.pdf

Reviewer 2 Report

In my opinion the work is interesting. For the purposes of reproducibility of the results, considering that the datasets are easily available on the web, provide more detailed information on the number of layers / neurons, optimization algorithm, etc.. In the work, if this information is already reported, it is not, in my opinion, easily identifiable. It would also be useful to provide information on the code used.

Author Response

Kindly see the attachment, thanks.

Author Response File: Author Response.pdf

Reviewer 3 Report

This paper presents an approach to anomaly detection in Internet of Vehicular networks based on Explainable Neural Networks.

The authors should take into consideration the following issues:

• Page 11, before subsection 3.2: a supplemental section or subsection should be added to explain how different deep learning or clustering techniques and XNN are used to attain the proposed objective. Also, explain how the anomalies were incorporated in the deep learning / clustering or XNN.
• Page 12, subsection 3.2.6: when explaining the clustering algorithm, you should clearly define attributes and classes.
• Page 12, eq. (2): explain the meaning of notations “m” an “p”: “m” does not appear in the summation; “p” appears in the summation as both a variable and the upper summation limit. Explain these anomalies!
• Page 12: define the significance of function f(x) in eqs. (3) and (4).
• Page 12, line 383: with the same notations as in eqs. (3) and (4); avoid words like “Sigma”, “Beta” or “Gamma”.
• Page 13, eq. (5): define quantities used in eq. (5).
• Page 14 and further: define the meaning of the horizontal axis in figures 8, 9, 12, 13, 14 and 17.

Author Response

Kindly see the attachment, Thanks.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The new version of the article is consistent in expression, present in detail and prioritized the evolution of methods for assessing possible interference and intrusion into the proper functioning of IoV software, which constantly ensures the optimal movement of vehicles in autonomous mode, through wireless control and guidance signals. from 5G infrastructures and I recommend to be published

Reviewer 2 Report

Accept in present form

Reviewer 3 Report

No more comments.