# Feature Analysis of Smart Shoe Sensors for Classification of Gait Patterns

^{1}

^{2}

^{3}

^{4}

^{5}

^{*}

^{†}

## Abstract

**:**

## 1. Introduction

## 2. Materials and Methods

#### 2.1. Data Acquisition

#### 2.2. Segmentation

#### 2.3. Feature Extraction

#### 2.4. Feature Selection

#### 2.5. Feature Reduction

#### 2.6. Classifier

#### 2.6.1. Random Forest

#### 2.6.2. K-Nearest Neighbor

#### 2.6.3. Logistic Regression

#### 2.6.4. Support Vector Machine

#### 2.7. Experiment Setting

#### 2.8. Evaluation Metrics

## 3. Results

#### 3.1. Feature Selection and Reduction

#### 3.2. Significance of Smart Shoes Sensors to Gait Pattern Analysis

#### 3.3. Optimal Number of Principal Components for the Classification Performances

#### 3.4. Performance Gait pattern Classification on Each Segmentation Type

## 4. Discussion

## 5. Conclusions

## Author Contributions

## Funding

## Acknowledgments

## Conflicts of Interest

## Abbreviations

Acc+Gyro | Accelerometer and Gyroscope |

FN | False negative |

FP | False Positive |

GEI | Gait energi image |

GRFs | Ground contact forces |

HMM | Hidden Markov model |

KNN | K-nearest neighbor |

KSOM | Kohonen self-organizing mapping |

LR | Logistic regression |

LSS | Lumbar spinal canal stenosis |

O-SVM | Optimized support vector machine |

PCA | Principal component analysis |

PCs | Principal components |

Pre+Acc | Pressure sensor and accelerometer |

Pre+Gyro | Pressure sensor and gyroscope |

RBF | Radial basis function |

RF | Random forest |

SVM | Support vector machine |

TP | True positive |

TN | True negative |

## Appendix A. Classification Performance Across 18 Participants Using Three Different Types of Sensors, Two Different Types of Sensors and Individual Types of Sensors

**Figure A1.**The classification performances of methods using a combination of three different types of sensors: pressure sensors, accelerometer and gyroscope: (

**a**) accuracy performance; (

**b**) precision performance; and (

**c**) recall performance.

**Figure A2.**The average accuracy performance over a combination of two types of sensors: (

**a**) accuracy over Pre+Acc; (

**b**) accuracy over Pre+Gyro; and (

**c**) accuracy over Acc+Gyro.

**Figure A3.**The average precision performance over a combination of two types of sensors: (

**a**) precision over Pre+Acc; (

**b**) precision over Pre+Gyro; and (

**c**) precision over Acc+Gyro.

**Figure A4.**The average recall performance over a combination of two types of sensors: (

**a**) recall over Pre+Acc; (

**b**) recall over Pre+Gyro; and (

**c**) recall over Acc+Gyro.

**Figure A5.**The average accuracy performance over individual types of sensors: (

**a**) accuracy over pressure sensor; (

**b**) accuracy over accelerometer; and (

**c**) accuracy over gyroscope.

**Figure A6.**The average of precision performance over individual types of sensors: (

**a**) precision over pressure sensor; (

**b**) precision over accelerometer; and (

**c**) precision over gyroscope.

**Figure A7.**The average of recall performance over individual types of sensors: (

**a**) recall over pressure sensor; (

**b**) recall over accelerometer; and (

**c**) recall over gyroscope.

## References

- Thomas, M.J.; Roddy, E.; Zhang, W.; Menz, H.B.; Hannan, M.T.; Peat, G.M. The population prevalence of foot and ankle pain in middle and old age: A systematic review. Pain
**2011**, 152, 2870–2880. [Google Scholar] [CrossRef] - Menz, H.B.; Dufour, A.B.; Riskowski, J.L.; Hillstrom, H.J.; Hannan, M.T. Association of Planus Foot Posture and Pronated Foot Function With Foot Pain: The Framingham Foot Study. Arthritis Care Res.
**2013**, 65, 1991–1999. [Google Scholar] [CrossRef] [Green Version] - Horwood, A.M.; Chockalingam, N. Defining excessive, over, or hyper-pronation: A quandary. Foot
**2017**, 31, 49–55. [Google Scholar] [CrossRef] - Abdulhay, E.; Arunkumar, N.; Narasimhan, K.; Vellaiappan, E.; Venkatraman, V. Gait and tremor investigation using machine learning techniques for the diagnosis of Parkinson disease. Future Gener. Comput. Syst.
**2018**, 83, 366–373. [Google Scholar] [CrossRef] - Vienne, A.; Barrois, R.P.; Buffat, S.; Ricard, D.; Vidal, P.P. Inertial sensors to assess gait quality in patients with neurological disorders: A systematic review of technical and analytical challenges. Front. Psychol.
**2017**, 8, 817. [Google Scholar] [CrossRef] [Green Version] - Lalli, P.; Chan, A.; Garven, A.; Midha, N.; Chan, C.; Brady, S.; Block, E.; Hu, B.; Toth, C. Increased gait variability in diabetes mellitus patients with neuropathic pain. J. Diabetes Complicat.
**2013**, 27, 248–254. [Google Scholar] [CrossRef] [PubMed] - Doheny, E.P.; Walsh, C.; Foran, T.; Greene, B.R.; Fan, C.W.; Cunningham, C.; Kenny, R.A. Falls classification using tri-axis accelerometers during the five-times-sit-to-stand test. Gait Posture
**2013**, 38, 1021–1025. [Google Scholar] [CrossRef] [PubMed] - Wang, K.; Redmond, S.J.; Lovell, N.H. Monitoring for elderly care: The role of wearable sensors in fall detection and fall prediction research. In Tele Medicine and Electronic Medicine; CRC Press: Boca Raton, FL, USA, 2016; pp. 619–651. [Google Scholar]
- Kwolek, B.; Kepski, M. Improving fall detection by the use of depth sensor and accelerometer. Neurocomputing
**2015**, 168, 637–645. [Google Scholar] [CrossRef] - Feng, J.; Wick, J.; Bompiani, E.; Aiona, M. Applications of gait analysis in pediatric orthopaedics. Curr. Orthop. Pract.
**2016**, 27, 455–464. [Google Scholar] [CrossRef] - Brunnekreef, J.J.; Van Uden, C.J.; van Moorsel, S.; Kooloos, J.G. Reliability of videotaped observational gait analysis in patients with orthopedic impairments. BMC Musculoskelet. Disord.
**2005**, 6, 17. [Google Scholar] [CrossRef] [Green Version] - Macri, F.; Marques, L.; Backer, R.; Santos, M.; Belangero, W. Validation of a standardised gait score to predict the healing of tibial fractures. J. Bone Jt. Surg. Br. Vol.
**2012**, 94, 544–548. [Google Scholar] [CrossRef] - Gurchiek, R.D.; Choquette, R.H.; Beynnon, B.D.; Slauterbeck, J.R.; Tourville, T.W.; Toth, M.J.; McGinnis, R.S. Open-source remote gait analysis: A post-surgery patient monitoring application. Sci. Rep.
**2019**, 9, 1–10. [Google Scholar] [CrossRef] - Junior, P.R.F. Gait Analysis Using Wearable Sensor Inertial in a Child with CP after Orthopedic Surgery: Case Report. Clin. Case Rep. Int. Physiother. Case Rep.
**2018**, 2, 1048. [Google Scholar] - Simon, S.R. Quantification of human motion: Gait analysis—benefits and limitations to its application to clinical problems. J. Biomech.
**2004**, 37, 1869–1880. [Google Scholar] [CrossRef] - Huang, H.; Kuiken, T.; Lipschutz, R. A Strategy for Identifying Locomotion Modes Using Surface Electromyography. IEEE Trans. Biomed. Eng.
**2009**, 56, 65–73. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Nazmi, N.; Rahman, M.A.A.; Yamamoto, S.I.; Ahmad, S.A. Walking gait event detection based on electromyography signals using artificial neural network. Biomed. Signal Process. Control
**2019**, 47, 334–343. [Google Scholar] [CrossRef] - Mazzetta, I.; Zampogna, A.; Suppa, A.; Gumiero, A.; Pessione, M.; Irrera, F. Wearable sensors system for an improved analysis of freezing of gait in Parkinson’s disease using electromyography and inertial signals. Sensors
**2019**, 19, 948. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Ngo, T.T.; Makihara, Y.; Nagahara, H.; Mukaigawa, Y.; Yagi, Y. Similar gait action recognition using an inertial sensor. Pattern Recognit.
**2015**. [Google Scholar] [CrossRef] - Gouwanda, D.; Gopalai, A.A.; Khoo, B.H. A low cost alternative to monitor human gait temporal parameters–wearable wireless gyroscope. IEEE Sens. J.
**2016**, 16, 9029–9035. [Google Scholar] [CrossRef] - Kim, S.Y.; Kwon, G.I. Gravity Removal and Vector Rotation Algorithm for Step counting using a 3-axis MEMS accelerometer. J. Korea Soc. Comput. Inf.
**2014**. [Google Scholar] [CrossRef] - Fujiwara, S.; Sato, S.; Sugawara, A.; Nishikawa, Y.; Koji, T.; Nishimura, Y.; Ogasawara, K. The Coefficient of Variation of Step Time Can Overestimate Gait Abnormality: Test-Retest Reliability of Gait-Related Parameters Obtained with a Tri-Axial Accelerometer in Healthy Subjects. Sensors
**2020**, 20, 577. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Buckley, C.; Micó-Amigo, M.E.; Dunne-Willows, M.; Godfrey, A.; Hickey, A.; Lord, S.; Rochester, L.; Del Din, S.; Moore, S.A. Gait Asymmetry Post-Stroke: Determining Valid and Reliable Methods Using a Single Accelerometer Located on the Trunk. Sensors
**2020**, 20, 37. [Google Scholar] [CrossRef] [Green Version] - Min, S.D.; Kwon, C.K. Step Counts and Posture Monitoring System using Insole Type Textile Capacitive Pressure Sensor for Smart Gait Analysis. J. Korea Soc. Comput. Inf.
**2012**. [Google Scholar] [CrossRef] - Lou, C.; Wang, S.; Liang, T.; Pang, C.; Huang, L.; Run, M.; Liu, X. A graphene-based flexible pressure sensor with applications to plantar pressure measurement and gait analysis. Materials
**2017**, 10, 1068. [Google Scholar] [CrossRef] - Park, S.W.; Das, P.S.; Park, J.Y. Development of wearable and flexible insole type capacitive pressure sensor for continuous gait signal analysis. Org. Electron.
**2018**, 53, 213–220. [Google Scholar] [CrossRef] - Mannini, A.; Trojaniello, D.; Cereatti, A.; Sabatini, A.M. A machine learning framework for gait classification using inertial sensors: Application to elderly, post-stroke and huntington’s disease patients. Sensors
**2016**, 16, 134. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Gao, J.; Gu, P.; Ren, Q.; Zhang, J.; Song, X. Abnormal Gait Recognition Algorithm Based on LSTM-CNN Fusion Network. IEEE Access
**2019**, 7, 163180–163190. [Google Scholar] [CrossRef] - Lee, S.S.; Choi, S.T.; Choi, S.I. Classification of gait type based on deep learning using various sensors with smart insole. Sensors
**2019**, 19, 1757. [Google Scholar] [CrossRef] [Green Version] - Song, F.; Guo, Z.; Mei, D. Feature selection using principal component analysis. In Proceedings of the 2010 International Conference on System Science, Engineering Design and Manufacturing Informatization, Yichang, China, 12–14 November 2010; Volume 1, pp. 27–30. [Google Scholar]
- Vaughan, C.L.; Davis, B.L.; O’Conner, J. Dynamics of Human Gait. In Dynamics of Human Gait, 2nd ed.; Vaughan, C., Ed.; Kiboho: Cape Towen, South Africa, 1992; Chapter 2; p. 140. [Google Scholar]
- McDonald, S.; Tavener, G. Pronation and supination of the foot: Confused terminology. Foot
**1999**, 9, 6–11. [Google Scholar] [CrossRef] - Aggarwal, A.; Gupta, R.; Agarwal, R. Design and Development of Integrated Insole System for Gait Analysis. In Proceedings of the 2018 Eleventh International Conference on Contemporary Computing (IC3), Noida, India, 2–4 August 2018; pp. 1–5. [Google Scholar] [CrossRef]
- Cho, T.; Sunarya, U.; Yeo, M.; Hwang, B.; Koo, Y.S. Deep-ACTINet: End-to-End Deep Learning Architecture for Automatic Sleep-Wake Detection Using Wrist Actigraphy. MDPI Electron.
**2019**, 8, 1461. [Google Scholar] [CrossRef] [Green Version] - Peker, M.; Arslan, A.; Şen, B.; Çelebi, F.V.; But, A. A novel hybrid method for determining the depth of anesthesia level: Combining ReliefF feature selection and random forest algorithm (ReliefF+RF). In Proceedings of the 2015 International Symposium on Innovations in Intelligent SysTems and Applications (INISTA), Madrid, Spain, 2–4 September 2015; pp. 1–8. [Google Scholar]
- Liu, X.; Tang, J. Mass Classification in Mammograms Using Selected Geometry and Texture Features, and a New SVM-Based Feature Selection Method. IEEE Syst. J.
**2014**, 8, 910–920. [Google Scholar] [CrossRef] - Lei, S. A Feature Selection Method Based on Information Gain and Genetic Algorithm. In Proceedings of the 2012 International Conference on Computer Science and Electronics Engineering, Hangzhou, China, 23–25 March 2012; Volume 2, pp. 355–358. [Google Scholar] [CrossRef]
- Zhou, Y.; Zhang, X.; Wang, J.; Gong, Y. Research on speaker feature dimension reduction based on CCA and PCA. In Proceedings of the 2010 International Conference on Wireless Communications & Signal Processing (WCSP), Suzhou, China, 21–23 October 2010; pp. 1–4. [Google Scholar] [CrossRef]
- Shi, T.; Horvath, S. Unsupervised Learning With Random Forest Predictors. J. Comput. Graph. Stat.
**2006**, 15, 118–138. [Google Scholar] [CrossRef] - Liaw, A.; Wiener, M. Classification and Regression by randomForest. R News
**2002**, 1609–3631. [Google Scholar] [CrossRef] - Raschka, S. Python Machine Learning; Packt Publishing: Birmingham, UK, 2015. [Google Scholar]
- Harrison, M. Machine Learning Pocket Reference; O’Reilly Media, Inc.: Newton, MA, USA, 2019. [Google Scholar]
- Begg, R.; Palaniswami, M.; Owen, B. Support Vector Machines for Automated Gait Classification. IEEE Trans. Biomed. Eng.
**2005**, 52, 828–838. [Google Scholar] [CrossRef] [PubMed] - Olson, D.L.; Delen, D. Advanced Data Mining Techniques; Number January; Springer: Berlin/Heidelberg, Germany, 2008; pp. 1–180. [Google Scholar] [CrossRef]
- Rehman, A.; Khan, A.; Ali, M.A.; Khan, M.U.; Khan, S.U.; Ali, L. Performance Analysis of PCA, Sparse PCA, Kernel PCA and Incremental PCA Algorithms for Heart Failure Prediction. In Proceedings of the 2020 International Conference on Electrical, Communication, and Computer Engineering (ICECCE), Istanbul, Turkey, 12–13 June 2020; pp. 1–5. [Google Scholar]
- Abdallah Bashir, A. Comparative study on classification performance between support vector machine and logistic regression. Int. J. Mach. Learn. Cybern.
**2012**, 4. [Google Scholar] [CrossRef] - Lakes, E.; Allen, K. Gait analysis methods for rodent models of arthritic disorders: Reviews and recommendations. Osteoarthr. Cartil.
**2016**, 24, 1837–1849. [Google Scholar] [CrossRef] [PubMed] [Green Version] - Domínguez-Morales, M.J.; Luna-Perejón, F.; Miró-Amarante, L.; Hernández-Velázquez, M.; Sevillano-Ramos, J.L. Smart footwear insole for recognition of foot pronation and supination using neural networks. Appl. Sci.
**2019**, 9, 3970. [Google Scholar] [CrossRef] [Green Version] - Jiang, X.; Zhang, Y.; Yang, Q.; Deng, B.; Wang, H. Millimeter-Wave Array Radar-Based Human Gait Recognition Using Multi-Channel Three-Dimensional Convolutional Neural Network. Sensors
**2020**, 20, 5466. [Google Scholar] [CrossRef] - Hayashi, H.; Toribatake, Y.; Murakami, H.; Yoneyama, T.; Watanabe, T.; Tsuchiya, H. Gait Analysis Using a Support Vector Machine for Lumbar Spinal Stenosis. Orthopedics
**2015**, 38, e959–e964. [Google Scholar] [CrossRef] [Green Version] - Mezghani, N.; Husse, S.; Boivin, K.; Turcot, K.; Aissaoui, R.; Hagemeister, N.; de Guise, J.A. Automatic Classification of Asymptomatic and Osteoarthritis Knee Gait Patterns Using Kinematic Data Features and the Nearest Neighbor Classifier. IEEE Trans. Biomed. Eng.
**2008**, 55, 1230–1232. [Google Scholar] [CrossRef] - Zeng, W.; Li, J.; Wang, Q.; Liu, F.; Wang, Y. Classification of gait patterns of anterior cruciate ligament deficient knees using gait analysis via deterministic learning. In Proceedings of the 2017 36th Chinese Control Conference (CCC), Dalian, China, 26–28 July 2017; pp. 10924–10929. [Google Scholar] [CrossRef]
- Zhang, D.; Wang, Y.; Bhanu, B. Age Classification Base on Gait Using HMM. In Proceedings of the 2010 20th International Conference on Pattern Recognition, Istanbul, Turkey, 23–26 August 2010; pp. 3834–3837. [Google Scholar] [CrossRef]
- Eskofier, B.; Lee, S.; Baron, M.; Simon, A.; Martindale, C.; Gaßner, H.; Klucken, J. An Overview of Smart Shoes in the Internet of Health Things: Gait and Mobility Assessment in Health Promotion and Disease Monitoring. Appl. Sci.
**2017**, 7, 986. [Google Scholar] [CrossRef] [Green Version] - Devi Das, K.; Saji, A.J.; Kumar, C.S. Frequency analysis of gait signals for detection of neurodegenerative diseases. In Proceedings of the 2017 International Conference on Circuit, Power and Computing Technologies (ICCPCT), Kollam, India, 20–21 April 2017; pp. 1–6. [Google Scholar] [CrossRef]
- Hsu, Y.; Chung, P.; Wang, W.; Pai, M.; Wang, C.; Lin, C.; Wu, H.; Wang, J. Gait and Balance Analysis for Patients With Alzheimer’s Disease Using an Inertial-Sensor-Based Wearable Instrument. IEEE J. Biomed. Health Inform.
**2014**, 18, 1822–1830. [Google Scholar] [CrossRef] [PubMed]

**Figure 1.**Block diagram for the identification of the gait patterns using pressure, three-axis accelerometer and three-axis gyroscope sensors. Four machine learning algorithms were applied—random forest, k-nearest neighbor, logistic regression and support vector machine—to classify five gait types.

**Figure 2.**The placement of sensors: eight pressure sensors were mounted on the outsole of the shoes as well as two Bosch BMI160 sensors. Each Bosch BMI160 sensor consists of a three-axis accelerometer and a three-axis gyroscope, respectively. There were 12 sensors in total mounted no the outsole of both the left and the right shoes. (

**a**) The position of sensors on the real board of smart shoe. (

**b**) The position of sensors outsole of the smart shoes.

**Figure 5.**Individual and cumulative explained variances for determining the optimal number of principal components. The cumulative explained variance shows the accumulation of variance for each principal component number. The individual explained variance describes the variance of each principal component.

**Figure 6.**The comparison of average accuracy on Acc+Gyro sensor combination using seven different segmentation types across 18 participants. The average accuracy was measured using SVM at optimal PCs number (68) on PCA.

No | Feature | Equation | Description |
---|---|---|---|

1 | Correlation | $Corr(x,y)={\displaystyle \frac{1}{N-1}}{\displaystyle \sum _{i=1}^{N}}{\displaystyle \frac{({x}_{i}-\overline{x})({y}_{i}-\overline{y})}{Std\left(x\right)Std\left(y\right)}}$ | Relation between two sensor channel data (x and y-axis). |

2 | Mean | $\overline{x}={\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^{N}}x\left(i\right)$ | Average of data x(i) with respect to data length of an epoch (N). |

3 | Standard Deviation | $Std\left(x\right)=\sqrt{{\displaystyle \frac{1}{N-1}}{\displaystyle \sum _{i=1}^{N}}{(x\left(i\right)-\overline{x})}^{2}}$ | Variation of a channel data. |

4 | Kurtosis | $K\left(x\right)={\displaystyle \frac{E{(x-\overline{x})}^{4}}{Std{\left(x\right)}^{2}}}-3$ | The tailedness of the probability distribution of one channel data using the 4th central moment with respect to variance [34]. |

5 | Crest Factor | $CF\left(x\right)={\displaystyle \frac{max\left(x\right(i\left)\right)}{\sqrt{{\displaystyle \frac{1}{N-1}}{\displaystyle \sum _{i=1}^{N}}x{\left(i\right)}^{2}}}}$ | How much extreme the peak of data is by measuring the ratio of the maximum value of the one sensor channel data to the effective value of the data. |

6 | Skewness | $\gamma \left(x\right)=E\left(\right)open="["\; close="]">{\left({\displaystyle \frac{x-\overline{x}}{Std\left(x\right)}}\right)}^{2}$ | An asymmetry measure of probability distribution of one channel data to its mean. |

7 | Entropy | $S\left(x\right)=-{\displaystyle \sum _{i}}{p}_{x}\left(i\right)ln\left({p}_{x}\left(i\right)\right)$ | Total probability mass function of one channel data [34]. |

8 | Spectral Flux | $SF\left(t\right)={\displaystyle \sum _{i=2}^{N}}{({x}_{t}\left(i\right)-{x}_{t-1}\left(i\right))}^{2}$ | The total difference between the successive data of one channel data. |

9 | Power | $P\left(x\right)={\displaystyle \frac{1}{N}}{\displaystyle \sum _{i=1}^{N}}{\left(x\right(i))}^{2}$ | The average energy of one channel data. |

Feature’s Name | Pressure Sensor | Accelerometer | Gyroscope |
---|---|---|---|

Correlation | 12 | 6 | 6 |

Mean | 8 | 6 | 6 |

Standard deviation | 8 | 6 | 6 |

Kurtosis | 8 | 6 | 6 |

Crest factor | 8 | 6 | 6 |

Skewness | 8 | 6 | 6 |

Entropy | 8 | 6 | 6 |

Spectral flux | 8 | 6 | 6 |

Power | 8 | $\phantom{\rule{4.2679pt}{0ex}}{6}^{-}$ | $\phantom{\rule{4.2679pt}{0ex}}{6}^{-}$ |

$Zerocrossin{g}^{-}$ | $\phantom{\rule{4.2679pt}{0ex}}{8}^{-}$ | $\phantom{\rule{4.2679pt}{0ex}}{6}^{-}$ | $\phantom{\rule{4.2679pt}{0ex}}{6}^{-}$ |

$Maxvalu{e}^{-}$ | $\phantom{\rule{4.2679pt}{0ex}}{8}^{-}$ | $\phantom{\rule{4.2679pt}{0ex}}{6}^{-}$ | $\phantom{\rule{4.2679pt}{0ex}}{6}^{-}$ |

**Table 3.**Accuracy of the gait pattern classification using the variance combinations of the smart shoes sensor features based on the optimal number of PCs. Pre, Acc and Gyro denote pressure sensor, accelerometer and gyroscope, respectively.

Sensor | Random Forest (%) | KNN (%) | L. Regression (%) | SVM (%) | PCs |
---|---|---|---|---|---|

Pressure | 62.56 $\pm \phantom{\rule{0.166667em}{0ex}}10.20$ | 51.89 $\pm \phantom{\rule{0.166667em}{0ex}}7.76$ | 35.76 $\pm \phantom{\rule{0.166667em}{0ex}}18.1$ | 65.03 $\pm \phantom{\rule{0.166667em}{0ex}}10.85$ | 33 |

Accelerometer | 53.81 $\pm \phantom{\rule{0.166667em}{0ex}}6.95$ | 45.21 $\pm \phantom{\rule{0.166667em}{0ex}}5.72$ | 50.06 $\pm \phantom{\rule{0.166667em}{0ex}}6.33$ | 61.19 $\pm \phantom{\rule{0.166667em}{0ex}}8.61$ | 33 |

Gyroscope | 53.63 $\pm \phantom{\rule{0.166667em}{0ex}}6.96$ | 46.45 $\pm \phantom{\rule{0.166667em}{0ex}}5.60$ | 42.99 $\pm \phantom{\rule{0.166667em}{0ex}}5.51$ | 60.44 $\pm \phantom{\rule{0.166667em}{0ex}}9.75$ | 34 |

Pre+Acc | 82.15 $\pm \phantom{\rule{0.166667em}{0ex}}9.54$ | 72.62 $\pm \phantom{\rule{0.166667em}{0ex}}9.76$ | 77.34 $\pm \phantom{\rule{0.166667em}{0ex}}11.52$ | 86.70 $\pm \phantom{\rule{0.166667em}{0ex}}8.15$ | 69 |

Pre+Gyro | 80.54 $\pm \phantom{\rule{0.166667em}{0ex}}10.32$ | 74.02 $\pm \phantom{\rule{0.166667em}{0ex}}10.54$ | 81.29 $\pm \phantom{\rule{0.166667em}{0ex}}10.09$ | 86.45 $\pm \phantom{\rule{0.166667em}{0ex}}9.41$ | 67 |

Acc+Gyro | 85.76$\pm \phantom{\rule{0.166667em}{0ex}}\mathbf{10.02}$ | 78.59$\pm \phantom{\rule{0.166667em}{0ex}}\mathbf{9.52}$ | 88.69$\pm \phantom{\rule{0.166667em}{0ex}}\mathbf{6.93}$ | 89.36$\pm \phantom{\rule{0.166667em}{0ex}}\mathbf{7.95}$ | 68 |

Pre+Acc+Gyro | 84.99 $\pm \phantom{\rule{0.166667em}{0ex}}9.72$ | 77.52 $\pm \phantom{\rule{0.166667em}{0ex}}10.23$ | 87.63 $\pm \phantom{\rule{0.166667em}{0ex}}7.52$ | 90.64 $\pm \phantom{\rule{0.166667em}{0ex}}6.98$ | 100 |

**Table 4.**Precision of the gait pattern classification using the variance combinations of the smart shoes sensor features based on the optimal number of principal components (PCs). Pre, Acc and Gyro denote pressure sensor, accelerometer and gyroscope, respectively.

Sensor | Random Forest (%) | KNN (%) | L. Regression (%) | SVM (%) | PCs |
---|---|---|---|---|---|

Pressure | 61.86 $\pm \phantom{\rule{0.166667em}{0ex}}11.34$ | 49.89 $\pm \phantom{\rule{0.166667em}{0ex}}7.82$ | 36.47 $\pm \phantom{\rule{0.166667em}{0ex}}17.30$ | 64.25 $\pm \phantom{\rule{0.166667em}{0ex}}10.88$ | 33 |

Accelerometer | 53.17 $\pm \phantom{\rule{0.166667em}{0ex}}11.73$ | 41.43 $\pm \phantom{\rule{0.166667em}{0ex}}6.59$ | 47.71 $\pm \phantom{\rule{0.166667em}{0ex}}9.45$ | 60.36 $\pm \phantom{\rule{0.166667em}{0ex}}9.83$ | 33 |

Gyroscope | 54.67 $\pm \phantom{\rule{0.166667em}{0ex}}10.06$ | 43.22 $\pm \phantom{\rule{0.166667em}{0ex}}5.73$ | 38.41 $\pm \phantom{\rule{0.166667em}{0ex}}7.34$ | 60.95 $\pm \phantom{\rule{0.166667em}{0ex}}10.54$ | 34 |

Pre+Acc | 84.54 $\pm \phantom{\rule{0.166667em}{0ex}}10.30$ | 72.86 $\pm \phantom{\rule{0.166667em}{0ex}}10.11$ | 78.33 $\pm \phantom{\rule{0.166667em}{0ex}}11.75$ | 87.09 $\pm \phantom{\rule{0.166667em}{0ex}}8.07$ | 69 |

Pre+Gyro | 81.71 $\pm \phantom{\rule{0.166667em}{0ex}}10.72$ | 74.50 $\pm \phantom{\rule{0.166667em}{0ex}}10.45$ | 81.49 $\pm \phantom{\rule{0.166667em}{0ex}}11.10$ | 86.36 $\pm \phantom{\rule{0.166667em}{0ex}}8.44$ | 67 |

Acc+Gyro | 87.04$\pm \phantom{\rule{0.166667em}{0ex}}\mathbf{9.94}$ | 78.16$\pm \phantom{\rule{0.166667em}{0ex}}\mathbf{10.35}$ | 88.30$\pm \phantom{\rule{0.166667em}{0ex}}\mathbf{7.44}$ | 89.76$\pm \phantom{\rule{0.166667em}{0ex}}\mathbf{8.11}$ | 68 |

Pre+Acc+Gyro | 86.41 $\pm \phantom{\rule{0.166667em}{0ex}}9.78$ | 78.23 $\pm \phantom{\rule{0.166667em}{0ex}}10.83$ | 87.64 $\pm \phantom{\rule{0.166667em}{0ex}}8.36$ | 91.08 $\pm \phantom{\rule{0.166667em}{0ex}}6.58$ | 100 |

**Table 5.**Recall of the gait pattern classification using the variance combinations of the smart shoes sensor features based on the optimal number of principal components (PCs). Pre, Acc and Gyro denote pressure sensor, accelerometer and gyroscope, respectively.

Sensor | Random Forest (%) | KNN (%) | L. Regression (%) | SVM (%) | PCs |
---|---|---|---|---|---|

Pressure | 59.35 $\pm \phantom{\rule{0.166667em}{0ex}}10.88$ | 50.05 $\pm \phantom{\rule{0.166667em}{0ex}}8.67$ | 35.02 $\pm \phantom{\rule{0.166667em}{0ex}}17.24$ | 64.47 $\pm \phantom{\rule{0.166667em}{0ex}}11.78$ | 33 |

Accelerometer | 44.55 $\pm \phantom{\rule{0.166667em}{0ex}}7.69$ | 41.50 $\pm \phantom{\rule{0.166667em}{0ex}}6.21$ | 41.99 $\pm \phantom{\rule{0.166667em}{0ex}}7.24$ | 56.73 $\pm \phantom{\rule{0.166667em}{0ex}}9.13$ | 33 |

Gyroscope | 40.64 $\pm \phantom{\rule{0.166667em}{0ex}}6.63$ | 43.62 $\pm \phantom{\rule{0.166667em}{0ex}}6.68$ | 33.26 $\pm \phantom{\rule{0.166667em}{0ex}}6.01$ | 55.68 $\pm \phantom{\rule{0.166667em}{0ex}}10.13$ | 34 |

Pre+Acc | 79.15 $\pm \phantom{\rule{0.166667em}{0ex}}10.86$ | 70.57 $\pm \phantom{\rule{0.166667em}{0ex}}10.78$ | 79.34 $\pm \phantom{\rule{0.166667em}{0ex}}11.36$ | 86.85 $\pm \phantom{\rule{0.166667em}{0ex}}8.62$ | 69 |

Pre+Gyro | 78.23 $\pm \phantom{\rule{0.166667em}{0ex}}11.77$ | 70.92 $\pm \phantom{\rule{0.166667em}{0ex}}12.11$ | 80.92 $\pm \phantom{\rule{0.166667em}{0ex}}10.54$ | 86.82 $\pm \phantom{\rule{0.166667em}{0ex}}10.41$ | 67 |

Acc+Gyro | 83.46 $\pm \phantom{\rule{0.166667em}{0ex}}11.36$ | 76.09 $\pm \phantom{\rule{0.166667em}{0ex}}11.11$ | 87.02 $\pm \phantom{\rule{0.166667em}{0ex}}7.05$ | 88.44 $\pm \phantom{\rule{0.166667em}{0ex}}8.46$ | 68 |

Pre+Acc+Gyro | 83.06 $\pm \phantom{\rule{0.166667em}{0ex}}11.18$ | 75.12 $\pm \phantom{\rule{0.166667em}{0ex}}11.47$ | 88.04 $\pm \phantom{\rule{0.166667em}{0ex}}7.59$ | 90.55 $\pm \phantom{\rule{0.166667em}{0ex}}7.15$ | 100 |

**Table 6.**Student t-test results which compare the classification performance of SVM with random forest, KNN and logistic regression.

Methods | Accuracy | Precision | Recall |
---|---|---|---|

Random Forest | $\phantom{\rule{6.25958pt}{0ex}}{0.008}^{\phantom{\rule{6.25958pt}{0ex}}**}$ | ${0.24}^{\phantom{\rule{6.25958pt}{0ex}}*}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.004}^{\phantom{\rule{6.25958pt}{0ex}}**}$ |

K-nearest neighbor | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{6.25958pt}{0ex}}**}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{6.25958pt}{0ex}}**}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{6.25958pt}{0ex}}**}$ |

Logistic regression | $0.397$ | $0.291$ | $0.520$ |

**Table 7.**The t-test of results performance between the combination of accelerometer-gyroscope sensors against other combinations and individual types of sensors on the SVM algorithm.

Sensor | Accuracy | Precision | Recall |
---|---|---|---|

Pressure | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ |

Accelerometer | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ |

Gyroscope | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ | $\phantom{\rule{6.25958pt}{0ex}}{0.001}^{\phantom{\rule{3.25958pt}{0ex}}**}$ |

**Table 8.**The effect of the number of PCs on classification performances of two and three different types of sensors.

Number of PCs | Acc+Gyro | Pre+Acc+Gyro | ||||
---|---|---|---|---|---|---|

Accuracy (%) | Precision (%) | Recall (%) | Accuracy (%) | Precision (%) | Recall (%) | |

10 | 77.43 | 78.08 | 77.21 | 78.10 | 78.04 | 78.29 |

30 | 86.26 | 86.45 | 85.44 | 86.13 | 86.50 | 86.31 |

50 | 88.72 | 89.01 | 87.88 | 88.25 | 88.50 | 88.39 |

60 | 89.06 | 89.56 | 88.30 | 88.37 | 88.69 | 88.28 |

68 ${}^{*}$ | 89.36 | 89.76 | 88.44 | 88.51 | 88.84 | 88.39 |

70 | 89.32 | 89.70 | 88.43 | 89.27 | 89.41 | 89.20 |

90 | 89.30 | 89.78 | 88.34 | 89.29 | 89.78 | 89.29 |

96 | 89.30 | 89.78 | 88.34 | 89.35 | 89.57 | 89.33 |

100 ${}^{*}$ | - | - | - | 90.64 | 91.08 | 90.55 |

120 | - | - | - | 90.70 | 91.30 | 90.70 |

140 | - | - | - | 90.71 | 91.37 | 90.61 |

160 | - | - | - | 90.51 | 91.35 | 89.93 |

172 | - | - | - | 90.49 | 91.32 | 89.90 |

Participants with the Worst Performances | Participants with the Best Performances | ||||||||||
---|---|---|---|---|---|---|---|---|---|---|---|

Participant 3 (average error rate of 25.4%) | Participant 5 (average error rate of 7.6%) | ||||||||||

Normal | 132 | 35 | 11 | 0 | 0 | Normal | 142 | 1 | 3 | 0 | 1 |

Left | 37 | 21 | 0 | 0 | 2 | Left | 1 | 48 | 0 | 1 | 0 |

Right | 19 | 0 | 37 | 0 | 0 | Right | 0 | 1 | 55 | 1 | 1 |

Toe-out-gait | 17 | 0 | 0 | 38 | 0 | Toe-out-gait | 11 | 3 | 0 | 37 | 0 |

Toe-in-gait | 0 | 3 | 0 | 0 | 50 | Toe-in-gait | 1 | 2 | 3 | 0 | 44 |

Error rate (%) | 35.6 | 64.4 | 22.9 | 0.0 | 3.8 | Error rate (%) | 8.4 | 12.7 | 9.8 | 2.6 | 4.3 |

Participant 8 (average error rate of 19.0%) | Participant 13 (average error rate of 0.9%) | ||||||||||

Normal | 123 | 5 | 2 | 3 | 10 | Normal | 140 | 1 | 0 | 1 | 0 |

Left | 23 | 23 | 0 | 1 | 0 | Left | 0 | 54 | 0 | 0 | 0 |

Right | 1 | 6 | 42 | 1 | 0 | Right | 0 | 0 | 52 | 0 | 0 |

Toe-out-gait | 1 | 3 | 0 | 46 | 0 | Toe-out-gait | 0 | 0 | 0 | 52 | 0 |

Toe-in-gait | 8 | 0 | 0 | 1 | 40 | Toe-in-gait | 1 | 0 | 0 | 0 | 56 |

Error rate (%) | 21.2 | 37.8 | 4.5 | 11.5 | 20. | Error rate (%) | 0.7 | 1.8 | 0 | 1.9 | 0 |

Participant 15 (average error rate of 32.3%) | Participant 18 (average error rate of 0.8%) | ||||||||||

Normal | 111 | 9 | 0 | 9 | 0 | Normal | 130 | 0 | 0 | 0 | 0 |

Left | 0 | 13 | 30 | 0 | 6 | Left | 1 | 53 | 0 | 0 | 0 |

Right | 0 | 0 | 49 | 0 | 0 | Right | 2 | 1 | 54 | 0 | 0 |

Toe-out-gait | 15 | 3 | 0 | 22 | 0 | Toe-out-gait | 0 | 0 | 0 | 52 | 0 |

Toe-in-gait | 1 | 2 | 0 | 0 | 14 | Toe-in-gait | 0 | 0 | 0 | 0 | 56 |

Error rate (%) | 12.6 | 51.9 | 38. | 29. | 30. | Error rate (%) | 2.3 | 1.9 | 0 | 0 | 0 |

Author | Number of Participants | Number of Classes | Method | Accuracy |
---|---|---|---|---|

Dominguez et al. [48] | 6 | 2: supination & pronation | NN | 90% |

Jiang et al. [49] | 8 | 2: jogging & walking | CNN | 92.5% |

Hayasi et al. [50] | 13 | 3: healthy, L4, & L5 | SVM | 84.6% |

Begg et al. [43] | 58 | 2: young & elderly | SVM | 90% |

2: asymptomatic, | Polynomial | 67% | ||

Mezghani et al. [51] | 42 | & osteoarthristis | representation | |

Wavelet | 91% | |||

Zeng et al. [52] | 46 | 2: healthy & anterior | RBF-NN | 93.47% |

cruciate ligament (ACL) | ||||

Counter+HMM | 83.33% | |||

Zhang et al. [53] | 14 | 2: old & young people | Silhouette+HMM | 76.24% |

Counter+Naive bayes | 65.85% | |||

Silhouette+Naive bayes | 63.28% | |||

Normal gait, unstable left, | ||||

This study | 18 | unstable right, supination, | SVM+PCA | 89.36% |

& pronation |

Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations. |

© 2020 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

**MDPI and ACS Style**

Sunarya, U.; Sun Hariyani, Y.; Cho, T.; Roh, J.; Hyeong, J.; Sohn, I.; Kim, S.; Park, C.
Feature Analysis of Smart Shoe Sensors for Classification of Gait Patterns. *Sensors* **2020**, *20*, 6253.
https://doi.org/10.3390/s20216253

**AMA Style**

Sunarya U, Sun Hariyani Y, Cho T, Roh J, Hyeong J, Sohn I, Kim S, Park C.
Feature Analysis of Smart Shoe Sensors for Classification of Gait Patterns. *Sensors*. 2020; 20(21):6253.
https://doi.org/10.3390/s20216253

**Chicago/Turabian Style**

Sunarya, Unang, Yuli Sun Hariyani, Taeheum Cho, Jongryun Roh, Joonho Hyeong, Illsoo Sohn, Sayup Kim, and Cheolsoo Park.
2020. "Feature Analysis of Smart Shoe Sensors for Classification of Gait Patterns" *Sensors* 20, no. 21: 6253.
https://doi.org/10.3390/s20216253