Smart Wearable to Prevent Injuries in Amateur Athletes in Squats Exercise by Using Lightweight Machine Learning Model

Round 1

Reviewer 1 Report

The idea of developing a smart wearable system to recognize squats is useful for analyzing and identifying bad postures in lifting sports. The authors performed various works, such as designing an integrated wearable hardware device and visual interface for signal acquisition and processing, and combining machine learning to find patterns in data. Moreover, the organization of the manuscript and its scientific discussions are well-explored. Therefore, I recommend publishing in the Information journal after addressing the below-mentioned queries:

1. Please avoid the abbreviation of “ML” in the title and abstract.

2. Page 1, Line 24-26, please supplement the wrong angle flexion picture of the main mistakes in Fig1.

3. Page 3, Part 3, Wearable design. It is recommended to supplement a block diagram of the system to introduce elements and relationships of electronic design.

4. Page 6, Line 181, “pre-processing techniques” is unclear.

5. Page 7, Line 191-193, why couldn’t the system recognize other main mistakes, such as wrong angle flexion?

6. Page 7, Line 195, what is the distribution of labels and experienced and inexperienced subjects in the dataset?

7. Page 6 and 8, Figure 4 and 5, change “a) 、b) 、c)”  to “A) 、B)、 C)” for ensuring the consistency of previous figures.

8. Page 9, Figure 6, the sensor is placed on one knee. Why not put sensors on both knees? Please analyze whether the result would be affected.

Note that the formatting of certain words or symbols needs to be standardized.

Author Response

Dear reviewer #1:

We, Ricardo Arciniega, Vanessa Erazo,  Paul Rosero, and Gyula Szabo, hereby present our manuscript entitled "Smart Wearable to Prevent Injuries in Amateur Athletes in Squats Exercise by Using Lightweight ML Model'' to be considered for publication as a regular track paper in MDPI Information.

 We are grateful for the provided comments and believe that the manuscript has been greatly improved based on your feedback. We have revised the manuscript based on your corrections and suggestions. We look forward to hearing whether you find this revision suitable for publication. The attached document shows our replies. 

Sincerely yours,

Ricardo Arciniega, PhD (c).

Author Response File: Author Response.pdf

Reviewer 2 Report

In the proposed work the authors are presenting a use of smart wearable devices for injuries prevention in squat exercises by using a lightweight machine-learning model. The motivation is behind possibility of injury because of erroneous squat movement, especially in amateur athletes’ population. The idea is to use a smart wearable device which can alert the athlete if performing of squats is not appropriate or even dangerous for injury. Data regarding correct squat movement was collected and models were trained with small memory demand, so they can be used in small embedded devices, like microcontrollers. K-nearest neighbours classification algorithm is used for classification. 85% accuracy performance is achieved with k=5. The main two contributions of the research work are novel machine-learning workflow running on the embedded device itself, and a comparison between classical machine-learning models and deep learning architectures for wearables.

The manuscripts is well-written, reads fluently, and the language is mostly good. Some minor errors/typos are present.

Chapter 2. Related work: a review of related work is very short. More detailed and extended review of the field should be made.  

Chapter 3. Wearable design: Regarding the description of the proposed embedded system, I think that detailed block diagram would be beneficial for better understanding of the device design. I encourage the authors to add block diagram of the design to the manuscript. Figure 4 provides some information, but it does not show how individual parts are interconnected.

Table 2: Temperature sensitivity -> What kind of property is this? Maybe you mean temperature operating range?

Regarding the Flex sensor authors did not mention, that the sensor has high resistance tolerance, which is +-30%!. This makes the Flex sensor more of an indicator than a sensor!

Line 133: MPU6050 is an accelerometer and a gyroscope. For determining the orientation, it uses accelerometer. I think this should be explained in more detail for a non-expert reader.

Authors state that the device consumes 157.9mA of current, but they do not say at which voltage. Probably at 3.3V.

ML workflow: It would be nice if some real data from the accelerometer + flex sensor would be graphically presented. Maybe a signal comparison between correctly and incorrectly executed squats could be made.

One concern: is a placement/positioning of the sensor system critical/crucial for the efficiency of the system? This should be addressed/discussed in the manuscript.

Line 73: … and ankle movements when athletes do swats. -> do squats.

Table 1: The load computational must be less as possible - …be as low as possible.

Line 136: …needs an analog-to-digital converter to process the incoming data.

Line 139: …how the sensor is blended. -> …how the sensor is bended.

Line 177: At the back part is placed the Sensor Flex 4.5", -> Please rewrite.

Line 254-255: using Tensorflow lite library to shirk the model and being able to run into the microcontroller. -> to shrink the model maybe???

Author Response

Dear reviewer #2:

We, Ricardo Arciniega, Vanessa Erazo,  Paul Rosero, and Gyula Szabo, hereby present our manuscript entitled ``Smart Wearable to Prevent Injuries in Amateur Athletes in Squats Exercise by Using Lightweight ML Model'' to be considered for publication as a regular track paper in MDPI Information.

 We are grateful for the provided comments and believe that the manuscript has been greatly improved based on your feedback. We have revised the manuscript based on your corrections and suggestions. We look forward to hearing whether you find this revision suitable for publication.  The attached document shows our replies.

Sincerely yours,

Ricardo Arciniega, PhD (c).

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Author have addressed all the comments and issues and gave sufficient reply.