Long Short-Term Memory (LSTM)-Based Dog Activity Detection Using Accelerometer and Gyroscope

Round 1

Reviewer 1 Report

Authors reported a dog activity detection and classification wearable device by using Long Short-Term Memory (LSTM) method. In order to solve the problem of dog health monitoring and comprehend their activity, this kind of wearable device is developed. The designed model in this article employed two sensors to overcome the disadvantage of LSTM, showing good experimental results with the accuracy of 94.25 % to collect 10 different activities. However, there are still some problems can be improved as follow.

1. The authors should present the advantages of LSTM comparing to other neural networks technology in Introduction part.

2. The authors can present more discussion for the different prediction outcome. For example, I can see the shape of line ‘running’ is different from the other lines, but without any explanation.

3. The clarity of Figure 3, 4 and 5 needs to be improved. 

4. The font of Figure 2 need to be adjusted again. It looks a little wider here.

Author Response

We would like to thank the Editor and reviewers for their in-depth reviews and constructive suggestions, which have substantially improved the quality of the manuscript. Many thanks to Reviewer#1 for the precious comment to further improve our paper.

Author Response File: Author Response.docx

Reviewer 2 Report

Joo et al. reported a long short-term memory (LSTM) model to classify dogs' activities using wearable sensors. This work is interesting for sensing technology, but there are many problems which should be addressed before its publication. 

1.      Is there a transition in dogs’ behavior from the short-term memory to long-term memory during the learning process? This key point should be investigated with necessary discussion. Meanwhile, does the proposed model have a better optimization space?

2.      For the introduction, some recently related advances in machine learning/neuromorphic computing and flexible/wearable sensors for monitoring activity are missing (i.e., The Innovation 2021, 2,100179; Mater. Today Electron. 2022, 1, 100001; Mater. Today Electron. 2022, 1, 100004). They are helpful to highlight the importance of artificial intelligence and wearable devices in the background.

3.      Figure 1 and Figure 2 are overlapping. The repeated Figure 2 should be removed from the manuscript.

4.      The quality of all figures needs to be improved. All formulas need to be numbered sequentially.

5.      There are several obvious defects/errors in the cited References, which should be checked and revised carefully.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Authors have developed a deep learning (long short-term memory (LSTM)) based algorithm to detect and classify ten types activities of dogs. Authors have collected data from ten trained dogs employing wearable devices with gyroscopes and accelerometers. Authors have used a low pass filter to remove the noise from the collected data. The authors have used the algorithm to analyze the filtered data to classify different activities of dogs. Over all, manuscript is well written and a detailed backgound information has been presented. However, following are to be addressed.

1. It is not clear how authors have determined true positive/negative and false positive/negative. Has video recording data been used for determining these?

2. Longer words of abbreviations AOC and ROC are not provided.

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The numerical order of all figures should be consecutive.

Author Response

Please see the attachment.

Author Response File: Author Response.docx