A Novel Feature Set Extraction Based on Accelerometer Sensor Data for Improving the Fall Detection System
Round 1
Reviewer 1 Report
The manuscript entitled: “A novel feature set extraction based on accelerometer sensor data for improving the fall-detection system” is relevant for the Electronics journal. The article is based on original experimental research. It should be emphasized that the presented results were described very precisely, and the attached drawings were prepared understandably. Overall, the paper is well prepared. Nevertheless, the article required small improvements:
- Line 16 - extra space
- Line 38 - why Kinect with a capital letter. Instead of saying "so on" it might be better to use etc.
- Line 110 - there should be a period instead of a comma at the end of the sentence.
- Line 139 - no spaces
- Line 149 - kNN is written once, then k-NN. Please standardize the notation
- Line 156-157 - "going" activity from small
- Line 279 - the phrase "in this study" is repeated many times, almost every chapter. As a rule, the purpose of the research and praising what was done in the article is presented at the end of the Introduction. Due to the article's own organization, such inserts are quite frequent, which makes it sometimes difficult to read.
- Line 298 - additional letter t
- Line 303 - MobiAct datasets were collected by Samsung Galaxy S3 sensors (premiere 2012). Why were newer technology smartphones not used? Maybe newer smartphones have more sensors, or they are more accurate and precise. Perhaps it would be appropriate to look for data from modern smartphones.
- Line 313 - MobiAct datasets were collected from Samsung Galaxy S3 sensors. Exactly what devices were used to collect the UP-Fall Detection (UP-Fall) dataset.
- Line 335-336 - "In this work ... Hjorth" - this sentence is not needed. Instead of using the word "each" in the next sentence, it might be better to mention the elements from the previous sentence.
Author Response
Response to Reviewer 1 Comments
Many thanks to the reviewer for their comments to improve the manuscript. We accept and have specific explanations for the opinions of the reviewer as follows:
Point 1:
- Line 16 - extra space
- Line 38 - why Kinect with a capital letter. Instead of saying "so on" it might be better to use etc.
- Line 110 - there should be a period instead of a comma at the end of the sentence.
- Line 139 - no spaces
- Line 149 - kNN is written once, then k-NN. Please standardize the notation
- Line 156-157 - "going" activity from small
- Line 279 - the phrase "in this study" is repeated many times, almost every chapter. As a rule, the purpose of the research and praising what was done in the article is presented at the end of the Introduction. Due to the article's own organization, such inserts are quite frequent, which makes it sometimes difficult to read.
- Line 298 - additional letter t
Response 1: We've fixed all typos, repeating the phrase "in this study" as suggested by reviewers.
Point 2: - Line 303 - MobiAct datasets were collected by Samsung Galaxy S3 sensors (premiere 2012). Why were newer technology smartphones not used? Maybe newer smartphones have more sensors, or they are more accurate and precise. Perhaps it would be appropriate to look for data from modern smartphones.
Response 2: MobiAct is a dataset collected and shared publicly by Vavoulas et al. [1] since 2016. It is based on the previously released MobiFall dataset (Vavoulas et al. 2014). At that time, Samsung Galaxy S3 is one of the most popular and sensors-packed smartphones. So they used it to collect data.
MobiAct is a data set used by many researchers on fall detection systems. This data set is collected methodically, scientifically, and accurately. So we also choose it for model building.
Point 3: - Line 313 - Exactly what devices were used to collect the UP-Fall Detection (UP-Fall) dataset.
Response 3: To collect data for the UP-Fall dataset, Martínez-Villaseñor et al. [2] used five Mbientlab MetaSensor wearable sensors collecting raw data from the 3-axis accelerometer, the 3-axis gyroscope, and the ambient light value. Also, one electroencephalograph (EEG) NeuroSky MindWave headset was occupied to measure the raw brainwave signal from its unique EEG channel sensor located at the forehead. They installed six infrared sensors as a grid 0.40 m above the floor of the room, to measure the changes in interruption of the optical devices. Lastly, two Microsoft LifeCam Cinema cameras were located at 1.82 m above the floor, one for a lateral view and the other for a frontal view.
Nevertheless, in this paper, we use the data obtained from the 3-axis accelerometer of the IMU device in the UP-Fall dataset. This device was placed in the right pocket of the volunteers' pants. The sampling frequency is standardized at 100Hz for all actions.
Point 4: - Line 335-336 - "In this work ... Hjorth" - this sentence is not needed. Instead of using the word "each" in the next sentence, it might be better to mention the elements from the previous sentence.
Response 4: We have revised this passage following the comments of the reviewer.
Author Response File: 
Author Response.pdf
Reviewer 2 Report
In this paper, the authors propose a data extraction method based on the time domain, frequency domain, and Hjorth parameter to build a dataset of 44 features of accelerometer data. They use two sets (MobiAct V2.0, UP-Fall) of accelerometer data with different collection methods to evaluate the effectiveness of the proposed method. The proposed dataset has also been tested on 05 different classifiers (SVM, kNN, J48, RF, and ANN algorithms) to confirm its superiority. The experiments illustrate that using a set combining 44-time domain, frequency domain, and Hjorth features allows increasing by 6.8%, 4.9%, 5.6%, 5.3%, 7.3% overall prediction accuracy (F1-measure) - in identifying various specific fall and non-fall activities for the MobileAct dataset when using SVM, kNN, J48, RF, and ANN algorithms respectively - as comparing to using a set containing only 41-time domain and frequency domain features. Some comments to improve the paper:
1) Some quantitative results need to be included in the abstract to show the outperformance of the proposed method compared to the state-of-the-art.
2) There are many dimensionality reduction methods that can select relevant features from multivariate data, such as KPCA, autoencoders. Please explain that.
3) Removing unnecessary information from the abstract by focusing on the problem statement, objectives, and findings.
4) Authors should provide the comments of the cited papers after introducing each relevant work. What readers require is, by convincing literature review, to understand the clear thinking/consideration why the proposed approach can reach more convincing results. This is the very contribution from the authors. In addition, authors also should provide more sufficient critical literature review to indicate the drawbacks of existing approaches, then, well define the mainstream of research direction, how did those previous studies perform? Employ which methodologies? Which problem still requires to be solved? Why is the proposed approach suitable to be used to solve the critical problem? We need more convinced literature reviews to indicate clearly the state-of-the-art development.
5) The methodology of the approach has to be more clearly presented.
6) Adding details about the reason for selecting that specific dataset in this study.
7) The complexity of the method needs to be provided.
8) The comparison with others should be included to show the benefit of the proposed approach.
9) Similar studies should be compared with the present research.
10) Limitations and future scope can be added.
11) A list of abbreviations should be added after the abstract.
Author Response
Response to Reviewer 2 Comments
Many thanks to the reviewer for their comments to improve the manuscript. We accept and have specific explanations for the opinions of the reviewer as follows:
Point 1: Some quantitative results need to be included in the abstract to show the outperformance of the proposed method compared to the state-of-the-art.
Response 1: We have added results to the abstract as suggested by reviewers.
Point 2: There are many dimensionality reduction methods that can select relevant features from multivariate data, such as KPCA, autoencoders. Please explain that.
Response 2: We focused on extracting the relevant features from raw data (the three-dimensional data collected from the accelerometer sensor). By applying some physical transformation based on the parameters Frequency domain, Time-domain, and Hjorth, we find the best presentation data for HAR, concentrated for detecting fall.
Point 3: Removing unnecessary information from the abstract by focusing on the problem statement, objectives, and findings.
Response 3: We have rewritten the abstract as suggested by the reviewer.
Point 4: Authors should provide the comments of the cited papers after introducing each relevant work. What readers require is, by convincing literature review, to understand the clear thinking/consideration why the proposed approach can reach more convincing results. This is the very contribution from the authors. In addition, authors also should provide more sufficient critical literature review to indicate the drawbacks of existing approaches, then, well define the mainstream of research direction, how did those previous studies perform? Employ which methodologies? Which problem still requires to be solved? Why is the proposed approach suitable to be used to solve the critical problem? We need more convinced literature reviews to indicate clearly the state-of-the-art development.
Response 4: We thank the reviewer for their valuable contributions. Although we have tried to present related studies carefully, details are in Sections 1 – Introduction and Section 2 – Related Work. We have re-read the citations and have added them to the article to make our proposed approach more convincing. We have added some relevant works to show the current status and convince.
Point 5: The methodology of the approach has to be more clearly presented.
Response 5: We have presented the approach of this study in Section 3.1.3 - The proposal Features extraction method. However, the reviewer considered the presentation approach to be unclear. We tried to rewrite it in common ways.
Point 6: Adding details about the reason for selecting that specific dataset in this study.
Response 6: We have added the reason for choosing two datasets, MobiAct V2.0 and UP-Fall for this study in Section 2 (Related Work). We have selected two open-access datasets, MobiAct v2.0 and UP-Fall, for feature extraction and FDS construction. These two data sets are collected methodically, scientifically, and diversely. The MobiAct v2.0 dataset has been collected from the accelerometer sensor of the smartphone. During the data collection process, volunteers were allowed to choose the location and orientation of the phone at random, so it was as close to reality as possible. The UP-Fall has been collected from the IMU (Inertial Measurement Unit) device. These are inexpensive and common devices in life. We use two data sets with different collection methods to ensure the stability of the results; reduce bias when evaluating system performance.
Point 7: The complexity of the method needs to be provided.
Response 7: In this study, our approach is to extract features from raw data collected from accelerometers of wearable sensors. So we don't focus on the complexity of machine learning models. We select classic machine learning models to test to evaluate the quality of the proposed feature set. The features in the frequency domain, in the time domain, and the Hjorth parameter are evaluated independently. Then combine the features of the domains together to improve recognition efficiency. Research shows that when combining features of all three domains Time, Frequency and Hjorth gives the best recognition performance. The proposed features are presented in Section 3.1.3. The effectiveness of the combination of features is evaluated by the experiment presented in Section 4.2.
Point 8: The comparison with others should be included to show the benefit of the proposed approach. Point 9: Similar studies should be compared with the present research.
Response 8 & 9: In this study, we compared the results with previous studies. Section 4.3.1 analyzes and evaluates advantages and disadvantages in research by Chatzaki et al[1]. This is a case study on Human Daily Activity and Fall Recognition based on the MobiAct V2.0 dataset. The experimental results based on our proposed method have been compared in detail with the study of Chatzaki et al. Section 4.3.2 is the same way as Section 4.3.1. We chose Lai[2] et al.'s study published in 2021 in the Pattern Recognition Letters journal for comparison. Research by Lai et al. obtained very high results, with careful analysis.
Point 10: Limitations and future scope can be added.
Response 10: We have added limitations and future scope to the paper as suggested by the reviewer.
Our proposed method can be extended for detecting abnormal activities beside falls among complex daily activities. Mobile applications for real-time fall detection and warning based on our model can be easily and feasibly implemented due to its low computing resource consumption.
Point 11: A list of abbreviations should be added after the abstract
Response 11: We checked and added abbreviations in parentheses when first used in the article. The journal template does not include a list of abbreviations. We have to follow the correct template of articles published by the journal.
[1] Chatzaki, C.; Pediaditis, M.; Vavoulas, G.; Tsiknakis, M. Human Daily Activity and Fall Recognition Using a Smartphone’s Acceleration Sensor. 2017.
[2] Lai, K.; Yanushkevich, S.N.; Shmerko, V.; Hou, M. Capturing Causality and Bias in Human Action Recognition. Pattern Recognition Letters 2021, 147, 164–171.
Author Response File: Author Response.pdf
