pith. sign in

arxiv: 1907.04788 · v1 · pith:IHMNJRWZnew · submitted 2019-07-05 · 📡 eess.SP · cs.LG

A Mobile Cloud Collaboration Fall Detection System Based on Ensemble Learning

Tong Wu , Yang Gu , Yiqiang Chen , Yunlong Xiao , Jiwei Wang This is my paper

Pith reviewed 2026-05-25 02:08 UTC · model grok-4.3

classification 📡 eess.SP cs.LG
keywords fall detectionensemble learningdecision treemobile cloudsensitivityspecificityactivities of daily livingthreshold filtering
0
0 comments X

The pith

A three-stage mobile-cloud system using an ensemble decision tree improves fall detection sensitivity and specificity by 1-3 percent over other methods.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper sets out to establish that splitting fall detection across a mobile device and the cloud produces more accurate results than algorithms that run entirely on one platform. It introduces a lightweight threshold filter on the phone to discard most everyday movements, then sends candidate signals to the cloud where an ensemble of decision trees performs the final classification. A sympathetic reader would care because falls remain a leading cause of injury, and a system that maintains high detection rates while limiting power use and data transfer could support continuous monitoring outside labs. The reported experiments indicate the approach works in real-world conditions rather than only controlled tests.

Core claim

The paper claims that the Falldetection Ensemble Decision Tree (FEDT), an ensemble learning method based on decision tree, deployed within a mobile cloud collaboration system divided into mobile-stage filtering of activities of daily living, collaboration-stage data transmission and cloud feature extraction, and cloud-stage classification, outperforms other methods by 1-3 percent on both sensitivity and specificity while delivering reliable fall detection in practical scenarios.

What carries the argument

The Falldetection Ensemble Decision Tree (FEDT) ensemble learning method based on decision tree, operating inside a three-stage mobile-cloud architecture that first applies a light-weighted threshold on the device to remove non-fall events.

If this is right

  • The mobile stage reduces unnecessary transmissions by discarding most activities of daily living via threshold.
  • Feature extraction moves to the cloud during the collaboration stage, allowing heavier computation.
  • The FEDT model supplies the final detection result after receiving the extracted features.
  • The overall system achieves 1-3 percent gains in sensitivity and specificity compared with other methods.
  • The architecture supports reliable fall detection outside laboratory conditions.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same staged filtering plus ensemble classification could be tested on other wearable sensor tasks such as gait analysis or seizure detection.
  • If the threshold filter is tuned more conservatively, the cloud model would receive noisier inputs and might require retraining to preserve accuracy.
  • Deployment in a full alerting pipeline would depend on whether the cloud response latency stays low enough for timely intervention.

Load-bearing premise

The light-weighted threshold method on the mobile device can reliably filter out activities of daily living without discarding actual falls.

What would settle it

A dataset or field trial in which the mobile threshold discards a measurable fraction of verified falls or forwards so many non-falls that the cloud-stage FEDT model cannot maintain the claimed sensitivity and specificity.

Figures

Figures reproduced from arXiv: 1907.04788 by Jiwei Wang, Tong Wu, Yang Gu, Yiqiang Chen, Yunlong Xiao.

Figure 1
Figure 1. Figure 1: The workflow of the system A. Mobile Stage: Threshold Method In the mobile stage, tri-axial accelerometer sensor data is collected as x, y, z. The threshold method is used to roughly filter out the suspicious falls. In the training set, the root mean square (RMS) of tri-axial accelerometer sensor data is employed to filter out the ADLs, shown in Eq. 1. The threshold is set by the statistics of the training… view at source ↗
Figure 2
Figure 2. Figure 2: RMS of fall data and ADL data [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Threshold method B. Collaboration Stage: Mobile Cloud Collaboration and Feature Extraction TABLE I. REPRESENTATIVE FEATURES AND DESCRIPTION Feature Calculate Formula and Description ݅ߨ2െ ൜݌ݔ݁௠ܽ෍ = ௞ܥ coefficient_fft ݉݇ ݊ ൠ ௡ିଵ ௠ୀ଴ , ݇ = 0, … , ݊െ 1 ௜ݐ෍ = ܧ energy_abs ଶ ௜ୀଵ,…,௡ |௜ݐെ ାଵ௜ݐ| ෍ = ݏ݄ܽ݊݃݁ܥ changes_absolute ௜ୀଵ,…,௡ିଵ energy_ration _by_chunks Calculate the sum of squares of chunk i out N chunks exp… view at source ↗
Figure 4
Figure 4. Figure 4: Application demo on Samsung watch gear s3 IV. EXPERIMENTS In this section, we will deploy sufficient experiments to show the effectiveness of the proposed FEDT. A. Datasets In order to evaluate the performance of the proposed FEDT, three open datasets: SisFall [23], MobiAct [24], MMsys [25], and a practical dataset are used. SisFall [23]: The SisFall dataset recorded 38 subjects’ activities which were comp… view at source ↗
Figure 5
Figure 5. Figure 5: Figure (a) is the seamentation stragegy of fall data. Figure (b) is the silding window mechanism of ADL data. When segmenting the ADL data, we use different window size in different dataset. In the SisFall, the size is 200 while in the MMsys it is set as 100. The size of MobiAct is 600, and the size of our practical dataset is 300. The amount of the fall and ADL samples after segmentation is shown in TABLE… view at source ↗
Figure 6
Figure 6. Figure 6: Figure (a) is the results on MMsys dataset. Figure (b) is the results on MobiAct dataset. Figure (c) is the results on SisFall dataset. Figure (d) is results on practical dataset Furthermore, we compare our method with some of the state-of-the-art methods on different datasets. To avoid the bias when rebuilding these methods, we evaluate these methods only on the datasets adopted in their papers. On MobiAc… view at source ↗
Figure 7
Figure 7. Figure 7: Results of robustness experiments The results report that the FEDT has superior robustness in sensitivity. As the FEDT is an ensemble algorithm, the detection result is determined by not only a classifier but the whole classifiers. Through the ensemble method, the model trained by FEDT algorithm has more transfer ability between different datasets. V. CONCLUSION In this paper, we propose a novel fall detec… view at source ↗
read the original abstract

Falls are one of the important causes of accidental or unintentional injury death worldwide. Therefore, this paper presents a reliable fall detection algorithm and a mobile cloud collaboration system for fall detection. The algorithm is an ensemble learning method based on decision tree, named Falldetection Ensemble Decision Tree (FEDT). The mobile cloud collaboration system can be divided into three stages: 1) mobile stage: use a light-weighted threshold method to filter out the activities of daily livings (ADLs), 2) collaboration stage: transmit data to cloud and meanwhile extract features in the cloud, 3) cloud stage: deploy the model trained by FEDT to give the final detection result with the extracted features. Experiments show that the performance of the proposed FEDT outperforms the others' over 1-3% both on sensitivity and specificity, and more importantly, the system can provide reliable fall detection in practical scenario.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

0 major / 3 minor

Summary. The manuscript proposes a three-stage mobile-cloud collaboration system for fall detection. Stage 1 uses a lightweight threshold method on the mobile device to filter out activities of daily living (ADLs). Stage 2 transmits candidate data to the cloud while extracting features. Stage 3 applies the proposed Falldetection Ensemble Decision Tree (FEDT) model for final classification. Experiments on the SisFall public dataset report that FEDT improves sensitivity and specificity by 1-3% over compared methods, with an explicit verification that the mobile filter discards zero actual falls.

Significance. If the reported performance gains and the zero-fall-loss verification hold, the work provides a concrete demonstration of an efficient edge-cloud pipeline that reduces mobile computation while maintaining high detection reliability. The use of a public dataset and the explicit filter check are positive elements that support reproducibility and practical applicability in wearable health monitoring.

minor comments (3)
  1. [Abstract] Abstract: the claimed 1-3% improvement is stated without naming the baseline methods or reporting exact sensitivity/specificity values; adding these would allow immediate assessment of the result.
  2. [Section 4 (or equivalent methods section)] The description of the FEDT ensemble construction (how individual decision trees are trained and combined) would benefit from a short pseudocode listing or explicit parameter settings to aid replication.
  3. [Experimental results section] Table or figure presenting the per-method sensitivity/specificity numbers should include the number of cross-validation folds and any statistical significance test used to support the 1-3% claim.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for the positive assessment of our work and the recommendation for minor revision. We appreciate the recognition of the reproducibility benefits from using the public SisFall dataset and the explicit verification of the mobile filter.

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents an empirical fall-detection pipeline (mobile threshold filter + cloud feature extraction + FEDT ensemble) evaluated on the public SisFall dataset with explicit threshold values and a reported zero-fall-discard check. No equations, derivations, fitted parameters renamed as predictions, or self-citation load-bearing steps exist. All performance numbers are experimental outcomes on held-out data rather than reductions to the paper's own inputs by construction. The argument is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

1 free parameters · 0 axioms · 0 invented entities

Abstract-only review prevents identification of specific fitted values or background axioms; the system implicitly depends on an unstated threshold for ADL filtering and on the assumption that the training data for FEDT is representative of real-world falls.

free parameters (1)
  • ADL filtering threshold
    Light-weighted threshold method used in mobile stage to discard daily activities; value not reported.

pith-pipeline@v0.9.0 · 5688 in / 1234 out tokens · 19859 ms · 2026-05-25T02:08:05.606065+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

  1. [1]

    Challenges, issues and trends in fall detection systems,

    R. Igual, C. Medrano and I. Plaza, “Challenges, issues and trends in fall detection systems,” Biomedical engineering online, 2013

    work page 2013

  2. [2]

    A survey on automatic fall detection in the context of ambient assisted living systems[J]

    Spasova V, Iliev I. A survey on automatic fall detection in the context of ambient assisted living systems[J]. International journal of advanced computer research, 2014, 4(1): 94

    work page 2014

  3. [3]

    Shen J, Chen Y, Shen Z, et al. A Two-Stage Incremental Update Method for Fall Detection with Wearable Device[C]//2018 IEEE SmartWorld, Ubiquitous Intelligence & Computing, Advanced & Trusted Computing, Scalable Computing & Communications, Cloud & Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/ UIC/ ATC/CBDCom/IOP/SCI)...

    work page 2018

  4. [4]

    A survey on vision-based human action recognition[J]

    Poppe R. A survey on vision-based human action recognition[J]. Image and vision computing, 2010, 28(6): 976-990

    work page 2010

  5. [5]

    RT-Fall: A real-time and contactless fall detection system with commodity WiFi devices[J]

    Wang H, Zhang D, Wang Y, et al. RT-Fall: A real-time and contactless fall detection system with commodity WiFi devices[J]. IEEE Transactions on Mobile Computing, 2017, 16(2): 511-526

    work page 2017

  6. [6]

    Nadee C, Chamnongthai K. Ultrasonic array sensors for monitoring of human fall detection[C]//2015 12th International Conference on Electrical Engineering/Electronics, Computer, Telecommunications and Information Technology (ECTI-CON). IEEE, 2015: 1-4. 93 83 88 90 78 80 82 84 86 88 90 92 94 Sensitivity Specificity Precision F1-score Percentage (%)

    work page 2015

  7. [7]

    Radar fall motion detection using deep learning[C]//2016 IEEE radar conference (RadarConf)

    Jokanovic B, Amin M, Ahmad F. Radar fall motion detection using deep learning[C]//2016 IEEE radar conference (RadarConf). IEEE, 2016: 1-6

    work page 2016

  8. [8]

    Activity recognition using low resolution infrared array sensor[C]//2015 IEEE International Conference on Communications (ICC)

    Mashiyama S, Hong J, Ohtsuki T. Activity recognition using low resolution infrared array sensor[C]//2015 IEEE International Conference on Communications (ICC). IEEE, 2015: 495-500

    work page 2015

  9. [9]

    He J, Z ha ng Z , Yu W . I nte rrup t-D riv e n F al l De te c tio n System Realized via a Kalman Filter and kNN Algorithm[C]//2018 IEEE SmartWorld, Ubiquitous Intelligence & Computing, Advanced & Trusted Computing, Scalable Computing & Communications, Cloud & Big Data Computing, Internet of People and Smart City Innovation (SmartWorld/SCALCOM/UIC/ATC/CBD...

    work page 2018

  10. [10]

    Vilarinho T, Farshchian B, Bajer D G, et al. A combined smartphone and smartwatch fall detection system[C]//2015 IEEE International Conference on Computer and Information Technology; Ubiquitous Computing and Communications; Dependable, Autonomic and Secure Computing; Pervasive Intelligence and Computing. IEEE, 2015: 1443-1448

    work page 2015

  11. [11]

    L e v e r a g i n g t w o - s t a g e weighted ELM for multimodal wearables based fall detection[M]//Proceedings of ELM-2014 Volume 2

    C h e n Z , C h e n Y , H u L , e t a l . L e v e r a g i n g t w o - s t a g e weighted ELM for multimodal wearables based fall detection[M]//Proceedings of ELM-2014 Volume 2. Springer, Cham, 2015: 161-168

    work page 2014

  12. [12]

    A smart phone-based pocket fall accident detection, positioning, and rescue system[J]

    Kau L J, Chen C S. A smart phone-based pocket fall accident detection, positioning, and rescue system[J]. IEEE journal of biomedical and health informatics, 2015, 19(1): 44-56

    work page 2015

  13. [13]

    Sun J, Wang Z, Pei B, et al. Fall Detection Using Plantar Inclinometer Sensor[C]//2015 IEEE 12th Intl Conf on Ubiquitous Intelligence and Computing and 2015 IEEE 12th Intl Conf on Autonomic and Trusted Computing and 2015 IEEE 15th Intl Conf on Scalable Computing and Communications and Its Associated Workshops (UIC-ATC-ScalCom). IEEE, 2015: 1692-1697

    work page 2015

  14. [14]

    A customized human fall detection system using omni-camera images and personal information[C]//1st Transdisciplinary Conference on Distributed Diagnosis and Home Healthcare, 2006

    Miaou S G, Sung P H, Huang C Y. A customized human fall detection system using omni-camera images and personal information[C]//1st Transdisciplinary Conference on Distributed Diagnosis and Home Healthcare, 2006. D2H2. IEEE, 2006: 39-42

    work page 2006

  15. [15]

    Fall detection with multiple cameras: An occlusion-resistant method based on 3-d silhouette vertical distribution[J]

    Auvinet E, Multon F, Saint-Arnaud A, et al. Fall detection with multiple cameras: An occlusion-resistant method based on 3-d silhouette vertical distribution[J]. IEEE transactions on information technology in biomedicine, 2011, 15(2): 290-300

    work page 2011

  16. [16]

    Depth camera based fall detection using human shape and movement[C]//2016 IEEE International Conference on Signal and Image Processing (ICSIP)

    Merrouche F, Baha N. Depth camera based fall detection using human shape and movement[C]//2016 IEEE International Conference on Signal and Image Processing (ICSIP). IEEE, 2016: 586-590

    work page 2016

  17. [17]

    Combining domain knowledge and machine learning for robust fall detection[J]

    Mirchevska V, Luštrek M, Gams M. Combining domain knowledge and machine learning for robust fall detection[J]. Expert Systems, 2014, 31(2): 163-175

    work page 2014

  18. [18]

    I nte grat i ng t he e nric he d feature with machine learning algorithms for human movement and fall detection[J]

    L i C, L i n M , Y a ng L T , e t a l. I nte grat i ng t he e nric he d feature with machine learning algorithms for human movement and fall detection[J]. The Journal of Supercomputing, 2014, 67(3): 854-865

    work page 2014

  19. [19]

    He J, Hu C, Li Y. An autonomous fall detection and alerting system based on mobile and ubiquitous computing[C]//2013 IEEE 10th International Conference on Ubiquitous Intelligence and Computing and 2013 IEEE 10th International Conference on Autonomic and Trusted Computing. IEEE, 2013: 539- 543

    work page 2013

  20. [20]

    Fall detection by wearable sensor and one-class SVM algorithm[M]//Intelligent computing in signal processing and pattern recognition

    Zhang T, Wang J, Xu L, et al. Fall detection by wearable sensor and one-class SVM algorithm[M]//Intelligent computing in signal processing and pattern recognition. Springer, Berlin, Heidelberg, 2006: 858-863

    work page 2006

  21. [21]

    A fall detection method based on acceleration data and hidden Markov model[C]//2016 IEEE International Conference on Signal and Image Processing (ICSIP)

    Cao H, Wu S, Zhou Z, et al. A fall detection method based on acceleration data and hidden Markov model[C]//2016 IEEE International Conference on Signal and Image Processing (ICSIP). IEEE, 2016: 684- 689

    work page 2016

  22. [22]

    Time Series FeatuRe Extraction on basis of Scalable Hypothesis tests (tsfresh–A Python package),

    M. Christ, N. Braun, J. Neuffer, AW. Kempa-Liehr “Time Series FeatuRe Extraction on basis of Scalable Hypothesis tests (tsfresh–A Python package),” Neurocomputing, 2018

    work page 2018

  23. [23]

    SisFall: A fall and movement dataset,

    A. Sucerquia, JD. López, JF. Vargas-Bonilla, “SisFall: A fall and movement dataset,” Sensors, vol. 17, pp. 198 2017

    work page 2017

  24. [24]

    he MobiAct Dataset: Recognition of Activities of Daily Living using Smartphones,

    G. Vavoulas, C. Chatzaki, T. Malliotakis, M. Pediaditis, et al, “he MobiAct Dataset: Recognition of Activities of Daily Living using Smartphones,” ICT4AgeingWell. 2016

    work page 2016

  25. [25]

    Data set for fall events and daily activities from inertial sensors

    O. Ojetola, E. Gaura, J. Brusey, “Data set for fall events and daily activities from inertial sensors”, Proceedings of the 6th ACM multimedia systems conference, 2015

    work page 2015

  26. [26]

    Human daily activity and fall recognition using a smartphone’s acceleration sensor,

    C. Chatzaki, M. Pediaditis, G. Vavoulas, et al. “Human daily activity and fall recognition using a smartphone’s acceleration sensor,” International Conference on Information and Communication Technologies for Ageing Well and e-Health, 2016

    work page 2016

  27. [27]

    Accurate fall detection using 3-axis accelerometer sensor and MLF algorithm,

    A. Jahanjoo, MN. Tahan, M. J. Rashti, “Accurate fall detection using 3-axis accelerometer sensor and MLF algorithm,” Pattern Recognition and Image Analysis, 2017 [28] P. Tsinganos, A. Skodras, “A smartphone-based fall detection system for the elderly”, mage and Signal Processing and Analysis (ISPA), 2017

    work page 2017

  28. [28]

    Towards online and personalized daily activity recognition, habit modeling, and anomaly detection for the solitary elderly through unobtrusive sensing,

    L. Meng, C. Miao, C. Leung, “Towards online and personalized daily activity recognition, habit modeling, and anomaly detection for the solitary elderly through unobtrusive sensing,” Multimedia Tools and Applications, vol. 76, pp. 10779-10779, 2017

    work page 2017

  29. [29]

    D e t e c t i n g u n s e e n f a l l s f r o m wearable devices using channel-wise ensemble of autoencoders,

    S S . K h a n , B . T a a t i , “ D e t e c t i n g u n s e e n f a l l s f r o m wearable devices using channel-wise ensemble of autoencoders,” Expert Systems with Applications, 2017

    work page 2017

  30. [30]

    Window-size impact on detection rate of wearable-sensor-based fall detection using supervised machine learning,

    IPES. Putra, R. Vesilo, “Window-size impact on detection rate of wearable-sensor-based fall detection using supervised machine learning,” IEEE Life Sciences Conference (LSC), 2017

    work page 2017

  31. [31]

    Real- life/real-time elderly fall detection with a triaxial accelerometer,

    A. Sucerquia, JD. López, JF. Vargas-Bonilla, “Real- life/real-time elderly fall detection with a triaxial accelerometer,” Sensors, vol. 18, 2018

    work page 2018

  32. [32]

    A Smartphone-Based System for Detecting Falls Using Anomaly Detection,

    V . C a r l e t t i , A . G r e c o , A . S a g g e s e , M . V e n t o , “ A Smartphone-Based System for Detecting Falls Using Anomaly Detection,” International Conference on Image Analysis and Processing, 2017

    work page 2017

  33. [33]

    An Efficient Design of a Machine Learning-Based Elderly Fall Detector,

    LP. Nguyen, M. Saleh, RLB. Jeannès, “An Efficient Design of a Machine Learning-Based Elderly Fall Detector,” International Conference on IoT Technologies for HealthCare, 2017

    work page 2017