Analysis of a Smartphone-Based Architecture with Multiple Mobility Sensors for Fall Detection with Supervised Learning
<p>Basic architecture of the testbed.</p> "> Figure 2
<p>Example of the distribution of the sensors worn by an experimental subject. The green arrow indicates the position of the smartphone. Red arrows correspond to the SensorTag modules, which are attached to the user’s body by means of elastic bands.</p> "> Figure 3
<p>Representation of the spatial reference system of the employed sensing devices (devices are firmly attached to the subjects’ body to guarantee that the reference system does not change during the experiments). (<b>a</b>) SensorTag (<b>b</b>) Smartphone.</p> "> Figure 4
<p>Typical Flow chart of a supervised learning classification algorithm.</p> "> Figure 5
<p>Example of the performance of the SVM algorithm for a two-dimensional space (samples characterized by two input features): (<b>a</b>) distribution of the training data on the two-dimensional space (<b>b</b>) creation of the hyperplane and classification decision for a certain test data.</p> "> Figure 6
<p>Operation of <span class="html-italic">k</span>-NN classifier: (<b>a</b>) the data under study is located among the training patterns aiming at finding the <span class="html-italic">k</span> nearest neighbor (in the example <span class="html-italic">k</span> = 5); (<b>b</b>) after detecting the <span class="html-italic">k</span> nearest samples, the new sample is classified considering the ‘majority vote’ (most common class) of its neighbors.</p> "> Figure 7
<p>Example of an operation of a decision-tree algorithm: (<b>a</b>) after the training phase, the branches and the decision rules the tree are configured; (<b>b</b>) during the test phase, the features of an unclassified sample are utilized to apply the decision rules and determine the sample class.</p> "> Figure 8
<p>Results for the SVM algorithm. Analysis of the residuals: (<b>a</b>) Normal probability plot (crosses: empirical data, dashed line: theoretical normal fit); (<b>b</b>) Values of the residuals versus the predicted (or fitted) values.</p> "> Figure 9
<p>Comparison of the means of the performance metric (<math display="inline"> <semantics> <mrow> <msqrt> <mrow> <mi>S</mi> <mi>p</mi> <mo>·</mo> <mi>S</mi> <mi>e</mi> </mrow> </msqrt> </mrow> </semantics> </math> or geometric mean of the sensitivity and the specificity) obtained with the SVM algorithm for all the possible combinations of input features: A = <math display="inline"> <semantics> <mrow> <msub> <mi>μ</mi> <mrow> <mi>S</mi> <mi>M</mi> <mi>V</mi> </mrow> </msub> </mrow> </semantics> </math>, B = <math display="inline"> <semantics> <mrow> <msub> <mi>A</mi> <mrow> <msub> <mi>w</mi> <mrow> <mi>d</mi> <mi>i</mi> <mi>f</mi> <mi>f</mi> <mrow> <mo>(</mo> <mrow> <mi>m</mi> <mi>a</mi> <mi>x</mi> </mrow> <mo>)</mo> </mrow> </mrow> </msub> </mrow> </msub> </mrow> </semantics> </math>, C = <math display="inline"> <semantics> <mrow> <msub> <mi>σ</mi> <mrow> <mi>S</mi> <mi>M</mi> <mi>V</mi> </mrow> </msub> </mrow> </semantics> </math>, D = <math display="inline"> <semantics> <mrow> <msub> <mi>μ</mi> <mi>θ</mi> </msub> </mrow> </semantics> </math>, E = <math display="inline"> <semantics> <mrow> <msub> <mi>μ</mi> <mrow> <mi>S</mi> <mi>M</mi> <msub> <mi>V</mi> <mrow> <mi>d</mi> <mi>i</mi> <mi>f</mi> <mi>f</mi> </mrow> </msub> </mrow> </msub> </mrow> </semantics> </math>, F = <math display="inline"> <semantics> <mrow> <msub> <mi>μ</mi> <mrow> <mi>A</mi> <mi>p</mi> </mrow> </msub> </mrow> </semantics> </math>. For each combination, the <span class="html-italic">y</span>-axis of the figure indicates with ‘1’ or ‘0’ whether the corresponding feature is considered (‘1’) or not (‘0’). The combination with the best results is indicated with a blue arrow.</p> "> Figure 10
<p>Results for the SVM algorithm. Analysis of the residuals: (<b>a</b>) Normal probability plot (crosses: empirical data, dashed line: theoretical normal fit); (<b>b</b>) Values of the residuals versus the predicted (or fitted) values.</p> "> Figure 11
<p>Comparison of the means of the performance metric (<math display="inline"> <semantics> <mrow> <msqrt> <mrow> <mi>S</mi> <mi>p</mi> <mo>·</mo> <mi>S</mi> <mi>e</mi> </mrow> </msqrt> </mrow> </semantics> </math> or geometric mean of the sensitivity and the specificity) obtained with the SVM algorithm for all the possible combinations and positions of the sensors. The positions are indicated as: P (Pocket), C (Chest), W (Waist), Wr (Wrist) or A (Ankle). For each combination, the y-axis of the figure indicates with ‘1’ or ‘0’ whether the corresponding sensor is considered (‘1’) or not (‘0’) for the detection decision. The combination with the best results is indicated with a blue arrow.</p> ">
Abstract
:1. Introduction
2. State of the Art on Wearable Fall Detection Systems and Multisensory Architectures
3. Description of the Experimental Testbed
- -
- To include the falls and ADLs that are typically considered in similar datasets in the literature (see [26] for a detailed analysis of these existing datasets). This typical ADLs encompass from basic operation (walking, sitting down) to daily activities which may cause very specific patterns of the acceleration measurements (applauding, going upstairs or downstairs).
- -
- To incorporate some ‘sporting activities’ (hopping, running), normally neglected by all these datasets, which may cause strong acceleration peaks that must be discriminated from those originated during a fall event.
4. Machine Learning Algorithms and Selection of the Input Features
4.1. Feature Extraction: Selection of the Input Statistics of the Machine Learning Algorithms
- -
- Mean Signal Magnitude Vector (), which describes the mean motion or agitation level of the body throughout the movement. This variable is computable as the mean module of the acceleration vector during the analysis interval:
- -
- The standard deviation of the Signal Magnitude Vector (), which informs about the variability of the accelerationThis parameter may be clearly affected by the presence of ‘valleys’ and ‘peaks’ in the evolution of the acceleration.
- -
- The sudden fluctuation of the mobility during a fall can be also described by the mean absolute difference () between consecutive samples of the acceleration module [40]:
- -
- As a fall occurrence usually implies a change in the orientation of the body, we also consider the mean rotation angle (), computable as [40]:
- -
- While the subject remains in an upright position, the effect of the gravity strongly determines the value of the acceleration component in the direction which is perpendicular to the floor plane. As a consequence, the inclination of the body caused by the falls normally provokes a remarkable modification of the acceleration components that defines the plane parallel to the floor when the subject is standing. Thus, to characterize this phenomenon, we utilize as a new feature the mean module () of these acceleration components ( and in the case of the SensorTag motes, and and for the smartphone, as it can be appreciated from the resting upright position depicted in Figure 2).
4.2. Employed Supervised Learning Classification Algorithm
4.2.1. Support Vector Machine (SVM)
4.2.2. k-Nearest Neighbors (k-NN)
4.2.3. Naive Bayes
4.2.4. Decision Tree
5. Result and Discussion
5.1. Analysis of the Impact of the Selection of the Acceleration-Based Features on the Fall Detection Performance
5.1.1. Results for Support Vector Machine (SVM) Algorithm
5.1.2. Results for the k-Nearest Neighbors (k-NN) Algorithm
5.1.3. Results for the Naïve Bayes Algorithm
5.1.4. Results for the Decision Tree Algorithm
5.1.5. Impact of the Election of the Input Characteristics: Summary of the Results
- -
- The use of a higher number of input characteristics does not necessarily correlate with an enhancement in the behavior of the decision algorithms. For the four strategies under study, no statistically significant improvement (in the geometric mean of the specificity and the sensitivity) is achieved by using more than four inputs. In fact, the performance of some machine learning strategies even deteriorates when the number of considered inputs increases.
- -
- The most frequent statistics in the combinations that yield the best performances of the algorithms are (which describes the variability of the acceleration module) and (which is linked to the changes in the perpendicularity of the body with respect to the floor plane). The use of (which identifies the presence of sudden and brusque changes of the acceleration module) also increases the effectiveness of all the algorithms (except for k-NN). Conversely, (which characterizes the mean variation of two consecutive samples of the acceleration module during a certain time window) is not included in any of the best combinations of statistics for any algorithm. In any case, the optimal election of the input characteristics clearly depends on the particularities of the employed algorithm. As it can be noted from Table 7, which summarizes the error and the relative impact of the six possible input characteristics on the algorithm’s performance, the importance of the election of each parameter strongly varies from one algorithm to another. A universal set of parameters cannot be proposed to characterize the mobility with independence of the underlying AI technique selected to identify the fall patterns. Thus, results indicate that input characteristics must be carefully designed and individualized for each detection policy.
- -
- SVM obtains the best results, notably outperforming the other algorithms (especially k-NN). The next subsection thoroughly investigates this comparison between the machine-learning strategies when other impacting factor (the selected position of the sensors) is considered.
5.2. Study of the Importance of the Sensor Position for the Decision of Machine-Learning Fall Detection Algorithms
5.2.1. Results for the SVM Algorithm
5.2.2. Results for the k-NN Algorithm
5.2.3. Results for the Naive Bayes Algorithm
5.2.4. Results for the Decision Tree Algorithm
5.2.5. Summary and Discussion of the Results
- -
- The particular location of the sensors on the user’s body has a noteworthy influence on the effectiveness of the fall detector with independence of the chosen algorithm.
- -
- The best results are always achieved when the detection algorithm employs the acceleration measurements captured on the chest (or trunk) or waist, the two points that are closest to the gravity center of the human body. This conclusion is coherent with the results of previous works [17,33,61,62,63,64,65,66] that compared the performance of the FDS when operating on two or more positions. In this regard, we cannot forget that ergonomics is a key aspect in the design of any wearable system. The analysis of the state-of-the art on FDSs performed by Thilo et al. in [67] has shown that all the aspects related to ergonomics are almost permanently ignored by the related literature. Thus, for example, these authors prove that the research on FDS prototypes rarely takes into account the opinion of the elderly (the main target of this type of emergency systems). In this vein, chest may constitute a quite unnatural place to locate a FDS as attaching or fixing a sensor to the chest most probably results in some discomfort to the user. Conversely, a belt on the waist (to which the sensors can be easily bonded or stitched) may introduce a more ergonomic alternative.
- -
- For all the algorithms, the worst results are associated to the use of the sensor at the ankle. This can be justified by the fact that the mobility of the ankle does not describe the global stability of the body. Hence, many ADLs can be misidentified as falls (and vice versa) if this location is used as a sensing point.
- -
- The consideration of a trouser pocket as a location for the sensor does not seem to introduce any improvement in the results. The outcome of the detection process is negatively affected if the algorithms take into account the measurements of the accelerometer embedded in a smartphone in the pocket. Pockets may provide users with a comfortable solution to carry small sensors. However, the variability of the typology of the pockets (width, height, ergonomics, internal space, attachment to the body) among the users may impact on the capability of this location to characterize the mobility of the body. Authors in [58] already showed that smartphone-based fall detectors are prone to errors if the device shifts within the pocket. Thus, the use of loose pockets to transport the smartphone may clearly undermine the reliability of this device to implement a FDS. In addition, the motion of the thigh during falls can likely be mistaken by that registered during the execution of certain types of ADLs. This bad behavior of this position forces to reconsider the role of the smartphone as a wearable sensor in fall detection systems. Although many studies have proposed its use in FDSs (taking advantage of its popularity and ease of programming), its effectiveness would be very limited unless it is transported in an uncomfortable (e.g., on a belt) or unnatural place (some works have even tested its effectiveness in FDSs when it is firmly attached to the chest). In the case of the smartphones employed in this study, the sampling frequency of its acceleration sensor was higher than that of the sensor motes (200 Hz vs. 20 Hz), so it can also be deduced that a higher sampling rate does not correlate with a better performance.
- -
- The simultaneous application of the algorithm to the signals sent by more than two sensors never increases the global performance metric (i.e., the possible increase of the specificity never compensates the loss of sensitivity caused by the fact that the fall must be detected in several points). As a consequence, the increase of the complexity of the body sensor network (which undoubtedly affects its ergonomics and cost) is not connected to a higher efficacy of the detection decision and should not be contemplated. The number of sensors is much less important than the election of the sensor position. A single-sensor architecture (with an accelerometer attached to the chest) seems to be enough to maximize the system effectiveness. The use of a second ‘backup’ sensor on the waist does not degrade the performance while it could help to avoid false positives in case of any anomaly suffered by the device on the chest. In this regard, the use of a smartwatch on the wrist as a more natural backup detection point could be also contemplated. In fact, the combined use of a smartwatch and a smartphone has been proved to outperform the effectiveness of the detection decision when compared to a “stand-alone” smartphone-based FDS [68,69,70]. Özdemir has shown in [17] that a fall detection algorithm based on the measurements captured by a sensor on the wrist may lead to a sensitivity higher than 97%.
- -
- If we compare the results achieved by the sensor on the waist (used as a single sensor) for the different algorithms, we can conclude that the best performance is attained by the SVM algorithm (although the confidence intervals obtained for the four machine learning strategies partly overlap). For the case of the SVM, the confidence interval of the performance metric is in the interval [0.956–0.999] with a mean value of 0.9775, which entails minimum mean values for the specificity and the sensitivity higher than 95.5%. In addition, SVM is the algorithm that produces the lowest error (14.162%) for the ANOVA analysis. This lower error reduces the possible impact of other possible factors that have not been considered (e.g., differences in the mobility patterns or physical characteristics of the subject). In any case, the impact of inter subject variability on the results should be studied in future studies. Thus, the ANOVA analysis should be developed taking into account the individual performance of the algorithms when applied to each subject separately. For that purpose, a longer dataset with a higher number of samples per subject is required in order to obtain a better characterization of the performance metrics (sensitivity and specificity) for each participant.
6. Conclusions
Acknowledgments
Author Contributions
Conflicts of Interest
References
- World Health Organization (WHO). Ageing and Health—Fact Sheet No 404 (September 2015). Available online: http://www.who.int/mediacentre/factsheets/fs404/en/ (accessed on 17 January 2018).
- World Health Organisation (WHO). Falls—Fact Sheet (Updated August 2017). Available online: http://www.who.int/mediacentre/factsheets/fs344/en/ (accessed on 17 January 2018).
- World Health Organization. Ageing & Life Course Unit. WHO Global Report on Falls Prevention in Older Age; World Health Organization: Geneva, Switzerland, 2008. [Google Scholar]
- Lord, S.R.; Sherrington, C.; Menz, H.B.; Close, J.C.T. Falls in older people: risk factors and strategies for prevention; Cambridge University Press: Cambridge, UK, 2007. [Google Scholar]
- Habib, M.A.; Mohktar, M.S.; Kamaruzzaman, S.B.; Lim, K.S.; Pin, T.M.; Ibrahim, F. Smartphone-based solutions for fall detection and prevention: Challenges and open issues. Sensors 2014, 14, 7181–7208. [Google Scholar] [CrossRef] [PubMed]
- Khan, S.S.; Hoey, J.R. Review of fall detection techniques: A data availability perspective. Med. Eng. Phys. 2016, 39, 12–22. [Google Scholar] [CrossRef] [PubMed]
- Board of Governors of the Federal Reserve System. Consumers and Mobile Financial Services 2016, March 2016; Board of Governors of the Federal Reserve System: Washington, DC, USA, 2016.
- Casilari, E.; Luque, R.; Morón, M. Analysis of android device-based solutions for fall detection. Sensors 2015, 15, 17827–17894. [Google Scholar] [CrossRef] [PubMed]
- Zhang, T.; Wang, J.; Liu, P.; Hou, J. Fall detection by embedding an accelerometer in cellphone and using KFD algorithm. Int. J. Comput. Sci. Netw. Secur. 2006, 6, 277–284. [Google Scholar]
- Soangra, R.; Lockhart, T.E. Agreement in gait speed from smartphone and stopwatch for five meter walk in laboratory and clinical environments. Biomed. Sci. Instrum. 2014, 50, 254–264. [Google Scholar] [PubMed]
- Casilari, E.; Santoyo-Ramón, J.A.; Cano-García, J.M. Analysis of a Smartphone-Based Architecture with Multiple Mobility Sensors for Fall Detection. PLoS ONE 2016, 11, e0168069. [Google Scholar] [CrossRef] [PubMed]
- Banos, O.; Damas, M.; Pomares, H.; Prieto, A.; Rojas, I. Daily living activity recognition based on statistical feature quality group selection. Expert Syst. Appl. 2012, 39, 8013–8021. [Google Scholar] [CrossRef]
- Zhang, L.; Liu, T.; Zhu, S.; Zhu, Z. Human Activity Recognition based on Triaxial Accelerometer. In Proceedings of the 2012 7th International Conference on Computing and Convergence Technology (ICCCT), Seoul, Korea, 3–5 December 2012; pp. 261–266. [Google Scholar]
- Wang, L.; Gu, T.; Tao, X.; Lu, J. A hierarchical approach to real-time activity recognition in body sensor networks. Pervasive Mob. Comput. 2012, 8, 115–130. [Google Scholar] [CrossRef]
- Huynh, Q.T.; Nguyen, U.D.; Tran, S.V.; Tran, B.Q. Optimum location for sensors in fall detection. In Proceedings of the International Conference on Green and Human Information Technology (ICGHIT), Hanoi, Vietnam, 27 February–1 March 2013. [Google Scholar]
- Huynh, Q.T.; Nguyen, U.D.; Irazabal, L.B.; Ghassemian, N.; Tran, B.Q. Optimization of an Accelerometer and Gyroscope-Based Fall Detection Algorithm. J. Sens. 2015, 2015, 452078. [Google Scholar] [CrossRef]
- Özdemir, A.; Turan, A. An Analysis on Sensor Locations of the Human Body for Wearable Fall Detection Devices: Principles and Practice. Sensors 2016, 16, 1161. [Google Scholar] [CrossRef] [PubMed]
- Igual, R.; Medrano, C.; Plaza, I. A comparison of public datasets for acceleration-based fall detection. Med. Eng. Phys. 2015, 37, 870–878. [Google Scholar] [CrossRef] [PubMed]
- Texas Instruments SimpleLinkTM. Bluetooth Low Energy/Multi-Standard SensorTag. Available online: http://www.ti.com/ww/en/wireless_connectivity/sensortag/ (accessed on 6 March 2017).
- Medrano, C.; Igual, R.; Plaza, I.; Castro, M. Detecting falls as novelties in acceleration patterns acquired with smartphones. PLoS ONE 2014, 9, e94811. [Google Scholar] [CrossRef] [PubMed]
- Fudickar, S.; Lindemann, A.; Schnor, B. Threshold-based Fall Detection on Smart Phones. In Proceedings of the 7th International Conference on Health Informatics (HEALTHINF’2014), Angers, France, 3–6 March 2014. [Google Scholar]
- Casilari, E.; Santoyo-Ramón, J.A. UMAFall: Fall Detection Dataset (Universidad de Malaga). Available online: https://figshare.com/articles/UMA_ADL_FALL_Dataset_zip/4214283 (accessed on 5 April 2018).
- Rawassizadeh, R.; Momeni, E.; Dobbins, C.; Gharibshah, J.; Pazzani, M. Scalable Daily Human Behavioral Pattern Mining from Multivariate Temporal Data. IEEE Trans. Knowl. Data Eng. 2016, 28, 3098–3112. [Google Scholar] [CrossRef]
- Figueiredo, I.N.; Leal, C.; Pinto, L.; Bolito, J.; Lemos, A. Exploring smartphone sensors for fall detection. mUX J. Mob. User Exp. 2016, 5, 2. [Google Scholar] [CrossRef]
- Min, C.; Kang, S.; Yoo, C.; Cha, J.; Choi, S.; Oh, Y.; Song, J. Exploring current practices for battery use and management of smartwatches. In Proceedings of the 2015 ACM International Symposium on Wearable Computers—ISWC ’15, Osaka, Japan, 7–11 September 2015; ACM Press: New York, NY, USA, 2015; pp. 11–18. [Google Scholar]
- Casilari, E.; Santoyo-Ramón, J.-A.; Cano-García, J.-M. Analysis of public datasets for wearable fall detection systems. Sensors 2017, 17, 1513. [Google Scholar] [CrossRef] [PubMed]
- Bagalà, F.; Becker, C.; Cappello, A.; Chiari, L.; Aminian, K.; Hausdorff, J.M.; Zijlstra, W.; Klenk, J. Evaluation of accelerometer-based fall detection algorithms on real-world falls. PLoS ONE 2012, 7, e37062. [Google Scholar] [CrossRef] [PubMed]
- Jämsä, T.; Kangas, M.; Vikman, I.; Nyberg, L.; Korpelainen, R. Fall detection in the older people: From laboratory to real-life. Proc. Est. Acad. Sci. 2014, 63, 341–345. [Google Scholar] [CrossRef]
- Kangas, M.; Vikman, I.; Nyberg, L.; Korpelainen, R.; Lindblom, J.; Jämsä, T. Comparison of real-life accidental falls in older people with experimental falls in middle-aged test subjects. Gait Posture 2012, 35, 500–505. [Google Scholar] [CrossRef] [PubMed]
- Kotsiantis, S.B.; Zaharakis, I.D.; Pintelas, P.E. Machine learning: A review of classification and combining techniques. Artif. Intell. Rev. 2006, 26, 159–190. [Google Scholar] [CrossRef]
- Liu, S.H.; Cheng, W.C. Fall detection with the support vector machine during scripted and continuous unscripted activities. Sensors 2012, 12, 12301–12316. [Google Scholar] [CrossRef] [PubMed]
- Yoshida, T.; Mizuno, F.; Hayasaka, T.; Tsubota, K.; Wada, S.; Yamaguchi, T. A wearable computer system for a detection and prevention of elderly users from falling. In Proceedings of the 12th International Conference on Biomedical Engineering (ICBME), Singapore, 7–10 December 2005; Magjarevic, R.R., Nagel, J.H., Eds.; Springer: Singapore, 2005. [Google Scholar]
- Kangas, M.; Konttila, A.; Winblad, I.; Jamsa, T. Determination of simple thresholds for accelerometry-based parameters for fall detection. In Proceedings of the 29th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBS 2007), Lyon, France, 22–26 August 2007; pp. 1367–1370. [Google Scholar]
- Ngu, A.; Wu, Y.; Zare, H.; Polican, A.; Yarbrough, B.; Yao, L. Fall Detection using Smartwatch Sensor Data with Accessor Architecture. In Proceedings of the International Conference, ICSH 2017, Hong Kong, China, 26–27 June 2017. [Google Scholar]
- Sucerquia, A.; López, J.D.; Vargas-Bonilla, F. Real-Life/Real-Time Elderly Fall Detection with a Triaxial Accelerometer. Preprints 2018. [Google Scholar] [CrossRef]
- Chen, J.; Kwong, K.; Chang, D.; Luk, J.; Bajcsy, R. Wearable sensors for reliable fall detection. In Proceedings of the IEEE Engineering in Medicine and Biology 27th Annual Conference (EMBC), Shanghai, China, 1–4 September 2005; pp. 3551–3554. [Google Scholar]
- Rodriguez, J.; Mercuri, M.; Karsmakers, P.; Soh, P.J.; Leroux, P.; Schreurs, D. Automatic fall detector based on sliding window principle. In Proceedings of the 34th WIC Symposium on Information Theory in the Benelux and the Third Joint WIC/IEEE SP Symposium on Information Theory and Signal Processing in the Benelux, Leuven, Belgium, 30–31 May 2013; pp. 215–219. [Google Scholar]
- Kozina, S.; Gjoreski, H.; Gams, M.; Luštrek, M. Efficient Activity Recognition and Fall Detection Using Accelerometers. In Evaluating AAL Systems Through Competitive Benchmarking; Botía, J.A., Álvarez-García, J.A., Barsocchi, P., Kaori, F., Riedel, T., Eds.; Springer: Berlin/Heidelberg, Germany, 2013; pp. 13–23. [Google Scholar]
- Lombardi, A.; Ferri, M.; Rescio, G.; Grassi, M.; Malcovati, P. Wearable wireless accelerometer with embedded fall-detection logic for multi-sensor ambient assisted living applications. In Proceedings of the 2009 IEEE Sensors, Christchurch, New Zealand, 25–28 October 2009; pp. 1967–1970. [Google Scholar]
- Chen, K.-H.; Yang, J.-J.; Jaw, F.-S. Accelerometer-based fall detection using feature extraction and support vector machine algorithms. Instrum. Sci. Technol. 2016, 44, 333–342. [Google Scholar] [CrossRef]
- Kerdegari, H.; Samsudin, K.; Ramli, A.R.; Mokaram, S. Evaluation of fall detection classification approaches. In Proceedings of the 4th International Conference on Intelligent and Advanced Systems (ICIAS 2012), Kuala Lumpur, Malaysia, 12–14 June 2012; Volume 1, pp. 131–136. [Google Scholar]
- Delahoz, Y.S.; Labrador, M.A. Survey on fall detection and fall prevention using wearable and external sensors. Sensors 2014, 14, 19806–19842. [Google Scholar] [CrossRef] [PubMed]
- Schwickert, L.; Becker, C.; Lindemann, U.; Maréchal, C.; Bourke, A.; Chiari, L.; Helbostad, J.L.; Zijlstra, W.; Aminian, K.; Todd, C. Fall detection with body-worn sensors: A systematic review. Z. Gerontol. Geriatr. 2013, 46, 706–719. [Google Scholar] [CrossRef] [PubMed]
- Igual, R.; Medrano, C.; Plaza, I. Challenges, issues and trends in fall detection systems. Biomed. Eng. Online 2013, 12, 66. [Google Scholar] [CrossRef] [PubMed]
- Mubashir, M.; Shao, L.; Seed, L. A survey on fall detection: Principles and approaches. Neurocomputing 2013, 100, 144–152. [Google Scholar] [CrossRef]
- Kotsiantis, S.B.; Zaharakis, I.; Pintelas, P. Supervised machine learning: A review of classification techniques. Emerg. Artif. Intell. Appl. Comput. Eng. 2007, 160, 3–24. [Google Scholar]
- Zhang, T.; Wang, J.; Xu, L.; Liu, P. Fall Detection by Wearable Sensor and One-Class SVM Algorithm. In Intelligent Computing in Signal Processing and Pattern Recognition; Springer: Berlin/Heidelberg, Germany, 2006; pp. 858–863. [Google Scholar]
- Salgado, P.; Afonso, P. Body Fall Detection with Kalman Filter and SVM. In Proceedings of the 11th Portuguese Conference on Automatic Control (CONTROLO’2014), Porto, Portugal, 21–23 July 2014; Springer: Cham, Switzerland, 2015; pp. 407–416. [Google Scholar]
- Wang, Y.; Wu, K.; Ni, L.M. WiFall: Device-Free Fall Detection by Wireless Networks. IEEE Trans. Mob. Comput. 2017, 16, 581–594. [Google Scholar] [CrossRef]
- Albert, M.V.; Kording, K.; Herrmann, M.; Jayaraman, A. Fall classification by machine learning using mobile phones. PLoS ONE 2012, 7, e36556. [Google Scholar] [CrossRef] [PubMed]
- Putra, I.P.E.S.; Vesilo, R. Window-size impact on detection rate of wearable-sensor-based fall detection using supervised machine learning. In Proceedings of the 2017 IEEE Life Sciences Conference (LSC), Sydney, NSW, Australia, 13–15 December 2017; pp. 21–26. [Google Scholar]
- Gunale, K.G.; Mukherji, P. Fall detection using k-nearest neighbor classification for patient monitoring. In Proceedings of the IEEE International Conference on Information Processing (ICIP 2015), Pune, India, 16–19 December 2015. [Google Scholar]
- Liu, C.-L.; Lee, C.-H.; Lin, P.-M. A fall detection system using k-nearest neighbor classifier. Expert Syst. Appl. 2010, 37, 7174–7181. [Google Scholar] [CrossRef]
- Genoud, D.; Cuendet, V.; Torrent, J. Soft Fall Detection Using Machine Learning in Wearable Devices. In Proceedings of the 2016 IEEE 30th International Conference on Advanced Information Networking and Applications (AINA), Crans-Montana, Switzerland, 23–25 March 2016; pp. 501–505. [Google Scholar]
- Yang, X.; Dinh, A.; Chen, L. A wearable real-time fall detector based on Naive Bayes classifier. In Proceedings of the 23rd Canadian Conference on Electrical and Computer Engineering (CCECE), Calgary, AB, Canada, 2–5 May 2010. [Google Scholar]
- Ojetola, O.; Gaura, E.I.; Brusey, J. Fall Detection with Wearable Sensors—Safe (Smart Fall Detection). In Proceedings of the 7th International Conference on Intelligent Environments; Nottingham, UK, 25–28 July 2011; pp. 318–321. [Google Scholar]
- Stone, E.E.; Skubic, M. Fall Detection in Homes of Older Adults Using the Microsoft Kinect. IEEE J. Biomed. Heal. Inform. 2015, 19, 290–301. [Google Scholar] [CrossRef] [PubMed]
- Aguiar, B.; Rocha, T.; Silva, J.; Sousa, I. Accelerometer-based fall detection for smartphones. In Proceedings of the IEEE International Symposium on Medical Measurements and Applications (MeMeA), Lisbon, Portugal, 11–12 June 2014; pp. 1–6. [Google Scholar]
- Montgomery, D. Design and Analysis of Experiments; John Wiley and Sons: New York, NY, USA, 2001; Volume 5. [Google Scholar]
- Jaeger, T.F. Categorical data analysis: Away from ANOVAs (transformation or not) and towards logit mixed models. J. Mem. Lang. 2008, 59, 434–446. [Google Scholar] [CrossRef] [PubMed]
- Fang, S.-H.; Liang, Y.-C.; Chiu, K.-M. Developing a mobile phone-based fall detection system on android platform. In Proceedings of the Computing, Communications and Applications Conference (ComComAp), Hong Kong, China, 11–13 January 2012; BenLetaief, K., Zhang, Q., Eds.; pp. 143–146. [Google Scholar]
- Dai, J.; Bai, X.; Yang, Z.; Shen, Z.; Xuan, D. Mobile phone-based pervasive fall detection. Pers. Ubiquitous Comput. 2010, 14, 633–643. [Google Scholar] [CrossRef]
- Dai, J.; Bai, X.; Yang, Z.; Shen, Z.; Xuan, D. PerFallD: A pervasive fall detection system using mobile phones. In Proceedings of the 8th IEEE International Conference on Pervasive Computing and Communications Workshops (PERCOM Workshops), Mannheim, Germany, 29 March–2 April 2010; Becker, C., Conti, M., Eds.; pp. 292–297. [Google Scholar]
- Gannapathy, V.R.; Ibrahim, A.; Zakaria, Z.B.; Othman, A.R.B.; Latiff, A.A. Zigbee-Based Smart Fall Detection and Notification System with Wearable Sensor (e-SAFE). Int. J. Res. Eng. Technol. 2013, 2, 337–344. [Google Scholar]
- Bourke, A.K.; O’Brien, J.V.; Lyons, G.M. Evaluation of a threshold-based tri-axial accelerometer fall detection algorithm. Gait Posture 2007, 26, 194–199. [Google Scholar] [CrossRef] [PubMed]
- Mao, L.; Liang, D.; Ning, Y.; Ma, Y.; Gao, X.; Zhao, G. Pre-impact and Impact Detection of Falls Using Built-In Tri-accelerometer of Smartphone. In Proceedings of the Third International Conference on Health Information Science (HIS 2014), Shenzhen, China, 22–23 April 2014; Springer: Cham, Switzerland, 2014; pp. 167–174. [Google Scholar]
- Thilo, F.J.S.; Hürlimann, B.; Hahn, S.; Bilger, S.; Schols, J.M.G.A.; Halfens, R.J.G. Involvement of older people in the development of fall detection systems: A scoping review. BMC Geriatr. 2016, 16, 42. [Google Scholar] [CrossRef] [PubMed]
- Vilarinho, T.; Farshchian, B.; Bajer, D.G.; Dahl, O.H.; Egge, I.; Hegdal, S.S.; Lones, A.; Slettevold, J.N.; Weggersen, S.M. A Combined Smartphone and Smartwatch Fall Detection System. In Proceedings of the 2015 IEEE International Conference on Computer and Information Technology; Ubiquitous Computing and Communications; Dependable, Autonomic and Secure Computing; Pervasive Intelligence and Computing (CIT/IUCC/DASC/PICOM), Liverpool, UK, 26–28 October 2015; pp. 1443–1448. [Google Scholar]
- Maglogiannis, I.; Ioannou, C.; Spyroglou, G.; Tsanakas, P. Fall Detection Using Commodity Smart Watch and Smart Phone. In Artificial Intelligence Applications and Innovations; Maglogiannis, I.L., Papadopoulos, H., Sioutas, S., Makris, C., Eds.; Springer: Berlin/Heidelberg, Germany, 2014; pp. 70–78. [Google Scholar]
- Casilari, E.; Oviedo-Jiménez, M.A. Automatic fall detection system based on the combined use of a smartphone and a smartwatch. PLoS ONE 2015, 10, e0140929. [Google Scholar] [CrossRef] [PubMed]
- Liu, J.; Lockhart, T.E. Development and Evaluation of a Prior-to-Impact Fall Event Detection Algorithm. IEEE Trans. Biomed. Eng. 2014, 61, 2135–2140. [Google Scholar] [PubMed]
Subject ID | Gender | Age | Height (cm) | Weight (kg) |
---|---|---|---|---|
Subject 1 | Female | 67 | 156 | 76 |
Subject 2 | Female | 22 | 167 | 63 |
Subject 3 | Male | 68 | 168 | 97 |
Subject 4 | Male | 27 | 173 | 90 |
Subject 5 | Male | 24 | 179 | 68 |
Subject 6 | Male | 24 | 175 | 79 |
Subject 7 | Male | 28 | 195 | 81 |
Subject 8 | Female | 22 | 167 | 57 |
Subject 9 | Male | 55 | 170 | 83 |
Subject 10 | Male | 19 | 178 | 68 |
Subject 11 | Male | 26 | 176 | 73 |
Subject 12 | Female | 51 | 155 | 55 |
Subject 13 | Female | 18 | 159 | 50 |
Subject 14 | Female | 22 | 164 | 52 |
Subject 15 | Male | 26 | 179 | 67 |
Subject 16 | Male | 21 | 173 | 77 |
Subject 17 | Female | 27 | 166 | 66 |
Subject 18 | Male | 24 | 177 | 66 |
Subject 19 | Female | 23 | 163 | 93 |
Mean value | 31.26 | 170.53 | 71.63 | |
Standard deviation | 15.98 | 9.54 | 13.52 | |
Median value | 24 | 170 | 68 |
Movement Type | No. of Executions | No. of Training Samples | No. of Test Samples (for the 6 Sub-Sets) | |||||
---|---|---|---|---|---|---|---|---|
Applauding | 42 | 14 | 4 | 4 | 5 | 5 | 5 | 5 |
Raising both arms | 43 | 14 | 5 | 5 | 4 | 5 | 5 | 5 |
Emulating a phone call | 46 | 14 | 5 | 5 | 5 | 5 | 6 | 6 |
Opening a door | 43 | 14 | 5 | 5 | 5 | 5 | 4 | 5 |
Sitting on a chair and getting up | 64 | 14 | 9 | 9 | 8 | 8 | 8 | 8 |
Walking | 63 | 14 | 8 | 9 | 8 | 8 | 8 | 8 |
Bending | 59 | 14 | 7 | 7 | 8 | 8 | 8 | 7 |
Hopping | 53 | 14 | 6 | 6 | 6 | 7 | 7 | 7 |
Lying down on/standing up from a bed | 57 | 14 | 7 | 7 | 7 | 7 | 7 | 8 |
Going upstairs and downstairs | 40 | 7 | 6 | 5 | 5 | 5 | 6 | 6 |
Jogging | 28 | 9 | 3 | 3 | 3 | 3 | 4 | 3 |
Forwards fall | 71 | 13 | 10 | 10 | 10 | 10 | 9 | 9 |
Backwards fall | 73 | 14 | 9 | 10 | 10 | 10 | 10 | 10 |
Lateral fall | 64 | 14 | 8 | 8 | 9 | 9 | 8 | 8 |
Total | 746 | 183 | 92 | 93 | 93 | 95 | 95 | 95 |
Feature | % |
---|---|
A | 3.332 |
B | 42.312 |
C | 14.971 |
D | 0.701 |
E | 0.207 |
F | 25.807 |
A&B | 0.785 |
B&F | 3.032 |
Error | 5.218 |
Feature | % |
---|---|
A | 0.954 |
B | 6.065 |
C | 21.329 |
D | 4.120 |
E | 1.334 |
F | 19.393 |
Error | 21.250 |
Feature | % |
---|---|
A | 0.977 |
B | 11.552 |
C | 7.472 |
D | 12.604 |
E | 2.143 |
F | 32.782 |
A&F | 1.668 |
B&D | 1.494 |
B&F | 4.067 |
C&F | 1.361 |
Error | 13.484 |
Feature | % |
---|---|
A | 3.709 |
B | 6.775 |
C | 41.181 |
D | 1.450 |
E | 1.158 |
F | 14.913 |
C&F | 2.178 |
D&E | 1.979 |
Error | 12.110 |
Error | |||||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|---|---|
SVM | ✓ | 3.332% | ✓ | 42.312% | ✓ | 14.971% | - | 0.701% | - | 0.207% | ✓ | 25.807% | 5.218% |
k-NN | ✓ | 0.954% | - | 6.065% | ✓ | 21.329% | ✓ | 4.120% | - | 1.334% | ✓ | 19.393% | 21.250% |
Naive Bayes | - | 0.977% | ✓ | 11.552% | ✓ | 7.472% | ✓ | 12.604% | - | 2.143% | ✓ | 32.782% | 13.484% |
Decision Tree | ✓ | 3.709% | ✓ | 6.775% | ✓ | 41.181% | - | 1.450% | - | 1.158% | ✓ | 14.913% | 12.110% |
Trouser Pocket | Chest | Waist | Wrist | Ankle | ||
---|---|---|---|---|---|---|
SVM Error = 14.162%. | ✓ | - | - | - | - | [0.741–0.783] |
- | ✓ | - | - | - | [0.956–0.999] | |
- | - | ✓ | - | - | [0.935–0.978] | |
- | - | - | ✓ | - | [0.890–0.933] | |
- | - | - | - | ✓ | [0.724–0.766] | |
- | ✓ | ✓ | - | - | [0.918–0.961] | |
k-NN Error = 22.857% | ✓ | - | - | - | - | [0.833–0.875] |
- | ✓ | - | - | - | [0.950–0.993] | |
- | - | ✓ | - | - | [0.950–0.992] | |
- | - | - | ✓ | - | [0.903–0.946] | |
- | - | - | - | ✓ | [0.797–0.840] | |
- | ✓ | ✓ | - | - | [0.926–0.969] | |
Naive Bayes Error = 16.967% | ✓ | - | - | - | - | [0.792–0.853] |
- | ✓ | - | - | - | [0.922–0.981] | |
- | - | ✓ | - | - | [0.903–0.963] | |
- | - | - | ✓ | - | [0.863–0.899] | |
- | - | - | - | ✓ | [0.614–0.675] | |
- | ✓ | ✓ | - | - | [0.863–0.923] | |
Decision Tree Error = 40.620% | ✓ | - | - | - | - | [0.787–0.840] |
- | ✓ | - | - | - | [0.938–0.991] | |
- | - | ✓ | - | - | [0.913–0.965] | |
- | - | - | ✓ | - | [0.890–0.943] | |
- | - | - | - | ✓ | [0.825–0.878] | |
- | ✓ | ✓ | - | - | [0.896–0.949] |
© 2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).
Share and Cite
Santoyo-Ramón, J.A.; Casilari, E.; Cano-García, J.M. Analysis of a Smartphone-Based Architecture with Multiple Mobility Sensors for Fall Detection with Supervised Learning. Sensors 2018, 18, 1155. https://doi.org/10.3390/s18041155
Santoyo-Ramón JA, Casilari E, Cano-García JM. Analysis of a Smartphone-Based Architecture with Multiple Mobility Sensors for Fall Detection with Supervised Learning. Sensors. 2018; 18(4):1155. https://doi.org/10.3390/s18041155
Chicago/Turabian StyleSantoyo-Ramón, José Antonio, Eduardo Casilari, and José Manuel Cano-García. 2018. "Analysis of a Smartphone-Based Architecture with Multiple Mobility Sensors for Fall Detection with Supervised Learning" Sensors 18, no. 4: 1155. https://doi.org/10.3390/s18041155