Improved Particle Filter in Machine Learning-Based BLE Fingerprinting Method to Reduce Indoor Location Estimation Errors
Round 1
Reviewer 1 Report
Comments and Suggestions for AuthorsThis article presents a method for indoor localization based on the intensity of the
received signal (RSSI) from BLE beacons utilizing fingerprinting technique.
The technique used is based on the k-Nearest Neighbors (k-NN) algorithm and the use of a particle filter.
My primary concern is about the advantage of the proposed method. There are other published works that obtain better results than those presented in this article, for example:
Zhuang Y, Yang J, Li Y, Qi L, El-Sheimy N. Smartphone-Based Indoor Localization with Bluetooth Low Energy Beacons. Sensors. 2016; 16(5):596. https://doi.org/10.3390/s16050596
Röbesaat J, Zhang P, Abdelaal M, Theel O. An Improved BLE Indoor Localization with Kalman-Based Fusion: An Experimental Study. Sensors. 2017; 17(5):951. https://doi.org/10.3390/s17050951
That show similar or better results in their own test environments.
Perhaps, it would be interesting to compare the authors' work with these papers to show the possible advantages that their work provides over them.
On the other hand, the mathematical notation used is sometimes confusing, for example, from equation 9 onwards.
Furthermore, the authors argue in the introduction (line 65) that their method reduces computational complexity by using a particle filter, but this assertion is not substantiated in the body of the paper.
Some minor comments:
- Phrase on line 56 is duplicated.
- Remove the arrows in Figure 1.
- Sodel --> Model (line 116)
- Reduce the size of Figure 4.
- Figure 4 is not referenced from the main body of the paper.
- Eq 3. Where does the number 1.4 come from?
- How do you check that a particle has penetrated a wall? Please explain.
- Where does Eq. 7 come from?
- Line 192, why 4 points are used (this is explained latter in the paper).
- Eq. 10 What are the limits of the integral?
- The legends on the axes of Figure 5 are missing.
- There are some question marks of Table 1.
- Line 246, Why are the other kernels discarded?
- Do the angle brackets in Eq. 16 mean dot product? Please clarify.
- Section 4.2.2 How were the experiments performed to conclude that 4 Neighbors are sufficient?
Author Response
Please see the attachment.
Author Response File: 
Author Response.docx
Reviewer 2 Report
Comments and Suggestions for AuthorsThe abstract is informative and well describes the article's contribution; however, there is too little information on the accuracy of the proposed method. The last sentence, “experiment results showed the effectiveness of the proposed system,” should be extended to add some more details on how the effectiveness has been improved and compare, e.g., the median error.
The figures are not captioned with sufficient details, and there are formatting errors on some figures. It is not sufficient to say that the “environment” is presented in Figure 8. The caption on Figure 10 does not provide information about what is meant by 119 and 179. The caption of Figure 12 is not informative enough and is impossible to read, as it is covered by the content of the figure.
The proposed indoor positioning method based on fingerprinting is evaluated on a very small dataset. Using just a single room is not enough to prove the validity of the method. The error variation is very high, which shows that the results are subject to high variability if the characteristics of the room change. The room which has been used contains furniture, which makes the results less representative, as any change in the location of the desks, etc., may lead to different results – this is partially visible in the spatial variability of the error presented in Figure 11. The proposed method should be evaluated on some datasets available in the literature (there are multiple such datasets available, e.g. on Kaggle) or at least in a few locations to ensure the results are reproducible.
Additionally, as the data are not published, it is impossible to reproduce the results of the proposed method by other researchers working on the indoor positioning systems.
Author Response
Please see the attachment.
Author Response File: Author Response.docx
Round 2
Reviewer 1 Report
Comments and Suggestions for AuthorsThe new version of the article has been improved from the previous version. From my point of view, the article can be accepted in its current state.
Author Response
Thank you very much for taking the time to review this manuscript. You have helped us a lot in improving our manuscript.
Reviewer 2 Report
Comments and Suggestions for AuthorsIt is very good that the data has been shared, but a more persistent method than a link to google drive should be used, preferably an open data repository with a DOI.
The results presented in table 4 are unrealistic. The error values are always greater or equal 0. So if the average error is equal 0 for k-NN, there were no other values than 0, so the variance of error is also 0. Another problem is that both the k-NN ans SVM show better accuracy (since the average error is smaller), why do we need a novel method?
Author Response
Please see the attachment.
Author Response File: 
Author Response.pdf
