The document presents a thesis on an intelligent gesture classification system empowered by support vector machine (SVM) that utilizes electromyography (EMG) signals for recognizing hand movements. The proposed model achieves a high accuracy of 99.9% in classifying gestures by analyzing data gathered from eight EMG sensors. The research advocates for future applications in real-time scenarios and improvements in gesture classification methodologies.

2. Electromyography Based Intelligence Gesture
Classification Empowered With Support Vector
Machine
Presented by: Thesis Supervisor:
Sajid Rasheed Dr. Muhammad Adnan Khan
Roll No. 20
Department of Computer Science & Information Technology
MINHAJ UNIVERSITY LAHORE HAMDARD CHOWK
TOWNSHIP LAHORE

3. Outline
 Introduction
 Electromyography (EMG)
 EMG Images
 Literature Review
 Problem statement
 Contribution
 Objectives
 Datasets
 Input / Output Variables
 Proposed Methodology
 -Proposed System Model
 - Support Vector Machine (SVM)
 - Math Behind SVM
 -Performance Evaluating Parameters
 Results Analysis
 - Training results
 - Testing and Validation results
 - Accuracy of Proposed Model
 - Compare of Proposed Model with Previous
Models
 Conclusion
 - Conclusion
 - Future Work

4. Introduction
 With the growing presence of computerized systems in our day to day lives, the importance
of Human-Computer Interface (HCI) system has been increased. HCI defines optimal usage
for storage, connectivity, and display capabilities used in the information flow. The
development of simple applications to understand movements of the body of the user and
transform these applications into commands of machine that has been of considerable
importance during the last years [1].
 Specific biological signals could be used for neuronal communication with devices, and
could be obtained from the particular organ, body, or cell network such as the system of
nervous.
 Reference[1]

5. Continue…
 The example of these approaches are Electromyogram (EMG), Electro-encephalogram
(EEG), and Electrooculogram (EOG). These methods are particularly important to people
with physical disabilities. There have been several attempts to use gesture-based EMG
signals for HCI growth.
 There is currently ongoing work on the processing EMG Signal and controllers in a variety
of fields, including the development of a graphical interface continuous EMG Signal
Classification to help disabled people use word processing applications and other personal
computer applications.
 The EMG tool can be designed for the identification of gestures based on the signal analysis
of different muscle groups in motion.

6.  Electromyography (EMG) is the study of electrical signals in the muscles, called myoelectric
operation, which are derived from the surface of the skin via sensors [2].
 Electromyography is a medical technique that measures the health status of the muscles and
nerve cells that regulate them or motor neurons. Muscle-acquired EMG signals permit
sophisticated methods for identification, decomposition, sorting, and classification.
 Different EMG signal analysis methodologies and techniques for the completion of this
objective include quick and accurate ways to grasp the signal and its existence
electromyographic signals.
Electromyography (EMG)
 Reference[2]

7. Images of Electromyography
 Reference[3,4]

8. Literature Review
• Lobov et al. [5]
 Work on the classification of hand gestures and implement it in dynamic gamming environment.
 Proposed model classified seven hand movement by using Artificial Neural Network (ANN).
 For data accusation, EMG Thalmic bracelet used which contained eight sensors.
 Proposed model achieved accuracy up to 91.5% using ANN.
• Alejandro et al. [6]
 An automated hand or wrist gesture identification system based on techniques of supervised machine
learning.
 Proposed model used an open-access collection of 36 subjects that included recordings of EMG
signals.
 Six hand gestures classified by using Convolutional Neural Network (CNN) and random forest
model and obtained accuracy up to 94.77% and 95.39% respectively.
 Reference[5,6]

9. Continue…
• Benalcázar et al. [7]
 Proposed a real-time hand classification method for the classification of hand movement.
 Discussed model used raw data from surface of eight EMG signal to measure the movement of
forearm.
 Model used K-Nearest Neighbors (KNN) classifier to identify five hand gesture without any feature
extraction method
 Proposed model show the best classification accuracy of 89.5% to recognize hand gestures.
• Bian et al. [8]
 discussed four classification systems are used to identify hand gestures relying on pattern recognition
of electromyographic (sEMG) surface signals.
 The results indicate that both accuracy and the preparation time of the model are outperformed by
the support vector machine. System classification accuracy is about as high as 92.25 percent
 Reference[7,8]

10. Problem statement
 Before many decades, Electromyography (EMG) has been used to find the problem of
muscles and nerve cells that control them, as well as used in computer science applications
to control the input devices such as a mouse, joystick, and home hold devices. However, the
use of EMG in controlling computing devices still a matter of discussion.
 A lot of work has been done by the researcher on EMG signals with some classification
methodologies like Fuzzy logic, artificial neural networks, Support Vector Machine
(SVM), k-nearest neighbor, etc. to recognized hand movement and other body parts.
 But the classification of hand gestures still a huge problem and unable to used
commercially for controlling devices based on hand movements.

11. Contributions
 Accuracy, efficiency, and generalizability are the major challenges of the existing gesture
classification systems. To overcome these limitations, an Intelligence Gesture Classification
System Empowered with Support Vector Machine (IGCS-SVM) proposed to recognize hand
movement.
 The proposed model extract features from the surface of EMG through eight EMG sensors
then support vector machine used to classify extracted features to recognize hand gestures.
The system communicates with controlling devices through the Internet of Things (IoT).
 SVM is a supervised data classification learning methodology and provide accurate results
that are better than others. That’s why the researcher has taken up this task in the shape of a
support vector machine technique to solve the classification problem.

12. Objectives
 To enables quantification of the gestures’ fidelity in a dynamic gaming environment.
 To reduce miss rate and mean square rate of intelligence gesture classification system.
 To improve the accuracy of intelligence gesture classification system empowered with
support vector machine.

13. Datasets
 The proposed model acquire dataset from internet that is publically available
on the website of UCI Machine Learning Repository [8] to classify hand
gestures. EMG used an MYO Thalmic bracelet to acquire data that was warned
by the user in his/her forearm.
 For the collection of data, 36 subjects participate that worn Thalmic bracelets
and perform seven basic gestures. Dataset contains ten attributes, one attribute
is time that record in a millisecond, other eight attributes contain eight EMG
channel to record the movement of gestures, and one attribute is class that
contain eight gestures.  Reference[8]

14. Input / Output Variables
Sr. No. Input / Output Variable Name
Input 1 Time (ms)
Input 2 Channel I
Input 3 Channel II
Input 4 Channel III
Input 5 Channel IV
Input 6 Channel V
Input 7 Channel VI
Input 8 Channel VII
Input 9 Channel VIII
Output 1 Class

15. Detail of Output Variable
Class Label of Gestures
0 unmarked data
1 hand at rest
2 hand clenched in a fist
3 wrist flexion
4 wrist extension
5 radial deviations
6 ulnar deviations
7 extended palm

16. Hand/Wrist gestures considered in the Dataset

17. Block Diagram of Proposed Model

18. Feature Extraction
• Mean Absolute Value
 Mean absolute value of an electromyography signal is determined by taking the absolute value of the
signal average. It is an estimate of the mean absolute signal xj value in the length of a segment j
which is W samples.
𝑀𝐴𝑉 =
1
w 𝑗=1
𝑤
| xj |, where j= 1,……, w - 1
• Root Mean Square Value
 Root mean square value for the surface of Electromyography can be calculated in the following
manners:
𝑅𝑀𝑆 =
1
𝑊 𝑗=1
𝑊
𝑥 𝑦2
 Where, 𝑥 𝑦 represents signals of Electromyography and W represents the length of signals.

19. Support Vector Machine Classifier
 Support Vector Machine (SVM) is a supervised machine learning technique that helps in
solving big data classification problems, it provide classification learning model and
algorithm.
 The purpose of SVM is to decide the ideal hyperplane that divides two classes of space
points. The hyperplane must satisfy the criterion to have a possible maximal distance from
both classes.

20. Mathematical Model
• As we know that the equation of the line is
y2 = ay1 + c (1)
Where ‘a’ is a slope of a line and ‘c’ is the intercept, therefore
ay1 − y2 + c = 0
• Let y = y1 , y2 and Z = a, −1 then above equation can be written as
zy + c = 0 (2)
This equation is derived from 2-dimensional vectors. But in fact, it also works for any number
of dimensions, equation 2 also known as hyper plane equation.
• The direction of a vector y = y1 , y2 is written as Z and is defined as
z =
y1
| y |
+
y2
| y |
(3)

21. Mathematical Model
• Length of Vector y calculated as
| y | = y1+
2
y2+
2
y3+
2
… … … . . yn
2
• The dot product for n − dimensional vectors can be computed as
z. y = i=1
n
ziyi (4)
Let
f = x (z . y + c)
If sign (f) > 0 then correctly classified and if sign (f) < 0 then incorrectly classified
• Given a dataset D, we compute f on a training dataset
fi = xi (z . y + c)
Then F which is called functional margin of the dataset
F = min
i=1,2,3,..…..,n
fi

22. Mathematical Model
 When comparing hyperplanes, the hyperplane with the largest F will be complimentary selected. Where F is
called the geometric margin of the dataset.
 Our objective is to find an optimal hyperplane, which means we need to find the values of z and c of the
optimal hyperplane.
 SVM optimization problem is case of constrained optimization problem, Lagrange multipliers are used to solve
it.
• Lagrangian function is
ℒ z, c, λ = (1/2) z. z −
i=1
n
λi [xi z. yi + c − 1]
With respect to z
𝛻zℒ z, c, λ = 𝑧 − i=1
n
λi xi yi = 0 (5)
With respect to c
𝛻cℒ z, c, λ = i=1
n
λi xi = 0 (6)

23. Mathematical Model
From two equations (5) and (6) we get
z = i=1
n
λ xi yi and i=1
n
λi xi = 0 (7)
 Equation (7) only find the optimal value of z that is dependent on λ , so the value of λ must
be find and value of c also need both z and λ.
• After substitute the value of z in Lagrangian function ℒ then we get
z λ , c =
i=1
n
λi −
1
2
i=1
n
k=1
n
λi λkxi xk yiyk
Above equation is dual optimization problem
thus
max
λ
i=1
n
λi −
1
2 i=1
n
k=1
n
λi λkxi xk yiyk (8)
Subject to constraint is λi ≥ 0 , i = 1 … . n , i=1
n
λi xi = 0

24. Mathematical Model
 Because the constraints have inequalities, so we extend the Lagrangian multipliers method to the
Karush-Kuhn-Tucker (KKT) conditions. The complementary condition of KKT states that
λi xi zi. y∗ + c − 1 = 0 (9)
y∗
is the optimal point.
λ is positive value otherwise, λ is equal to 0 on other points
So
xi zi. y∗
+ c − 1 = 0 (10)
• These are called support vectors, which are the closest points to the hyperplane. According to the
above equation (10)
z −
i=1
n
λi xi yi = 0
z = i=1
n
λi xi yi (11)

25. Mathematical Model
• To calculate the value of c we find
xi zi. y∗ + c − 1 = 0 (12)
• In equation (12) multiply by x on both sides so we get
xi
2
zi. y∗
+ c − xi = 0
Where xi
2
= 1
zi. y∗ + c − xi = 0
c = x − zi. y∗ (13)
Then
c =
1
v i=1
v
( x − z . y) (14)
V is the number of support vectors. On one occasion we will have the hyperplane, then
we can use the hyperplane to make predictions.

26. Mathematical Model
• Where the hypothesis function is
h zi =
+1 if z. y + c ≥ 0
−1 if z. y + c < 0
(15)
 The above-mentioned point on the hyperplane is categorized as class + 1 (gesture
successfully classified) and the point below the hyperplane is categorized as class -1 (gesture
not classified).
 So, basically the goal of the SVM Algorithm is to find a hyperplane which could separate
the data accurately and we need to find the best one, which is often referred as the optimal
hyperplane.

27. Performance Evaluating Parameters
The objective/quantitative method includes performance evaluating metrics that
gives the statistical results. The quantitative way of assessment includes
 Accuracy
 Miss rate

28. • Accuracy can be defined as the percentage of correctly classified instances.
o Accuracy = (correctly predicted class / total testing class) × 100%.
𝐀𝐜𝐜 =
𝑻𝑷+𝑻𝑵
𝑻𝑷+𝑭𝑷+𝑻𝑵+𝑭𝑵
where TP, FN, FP and TN represent the number of true positives, false negatives, false
positives and true negatives, respectively.
• Miss Rate can be defined as the percentage of wrongly classified instances.
o Miss Rate = (wrongly predicted class / total testing class) × 100%.
Miss Rate=
𝑭𝑷+𝑭𝑵
𝑻𝑷+𝑭𝑷+𝑻𝑵+𝑭𝑵
Continued…

29. Results Analysis
 Training accuracy of proposed IGCS-SVM model in the form of number of
observation as well as Positive Predictive Value And False Discovery Rate
 Training section of proposed model contained 80% data of whole dataset which
contain 8669 samples to predict seven classes of hand gestures.
Training Results of Proposed Model

30. Number Of Observation In Training Phase

31. Positive Predictive Value And False Discovery Rate In Training
Phase

32. Testing and Validation Phase
 Testing and Validation accuracy of proposed IGCS-SVM model in the form of
number of observation as well as Positive Predictive Value And False Discovery
Rate.
 Testing and Validation phase of proposed model contained 20% data of whole
dataset which contain 2168 samples to predict seven classes of hand gestures.

33. Number Of Observation In Testing and Validation Phase

34. Positive Predictive Value And False Discovery Rate In Testing &
Validation Phase

35. Training And Validation Accuracy of Proposed Model
Accuracy Miss rate
Training 99.2% 0.8%
Validation 99.9% 0.1%

36. Comparison of Proposed IGCS-SVM With Previous Work
Model Accuracy Miss Rate
Benalcazar et al. (2017) [10] 86% 14%
Chawathe (2019) [9] 89% 11%
Lobov et al. (2018) [5] 91.5% 8.5%
Alejandro et al. (2020) [6]
CNN Model
94.77% 5.23%
Random Forest Model 95.39% 4.61%
Proposed IGCS-SVM Model 99.9% 0.1%  Reference[5,6,9,10]

37. Conclusion
 In current thesis, an IGCS-SVM model is proposed for intelligent gesture classification
system based on Electromyography (EMG) signals.
 Proposed model collect data from eight EMG sensors and then analyze it to classified
gesture. Support vector machine classified hand gestures in this model.
 EMG signals acquired from different muscles location, through the Mayo armband Thalmic
bracelet, then support vector machine classified acquired signals. The proposed model
communicates with computing devices through IoT.
 Presented IGCS-SVM model achieved gesture classification accuracy 99.9% using SVM.
Computational results show that the support vector machine proved a good choice to classify
hand gestures.
 The simulation findings show that the suggested methodology produced batter outcomes as
compared to the previous approaches used by model Lobov et al. (2018) [5], Alejandro et al
(2020) [6], Chawathe (2019)[9] and Benalcazar et al. (2017) [10].

38. Future Work
 The present research opened up innovative opportunities for future researchers in the area of
human-computer interaction by implementing the efficiencies of the proposed Intelligence
Gesture Classification System empowered with Support Vector Machine (IGCS-SVM.
 In the future, a real-time application build using this technique. Furthermore, we will use
new classification algorithms for classification and feature extraction to build models that
enhance the performance of the real-time application.

39. [1]. Ahsan, M. R., Ibrahimy, M. I., & Khalifa, O. O. (2009). EMG signal classification for human computer interaction: a
review. European Journal of Scientific Research, 33(3), 480- 501.
[2]. Reaz, M. B. I., Hussain, M. S., & Mohd-Yasin, F. (2006). Techniques of EMG signal analysis: detection, processing,
classification and applications. Biological procedures online, 8(1), 11-35.
[3]. https://encrypted-tbn0.gstatic.com/images?q=tbn%3AANd9GcQAApwTeIx8t4NV9Yz5kA71grP2wepYR1-p8g&usqp=CAU
[4]. https://www.mdpi.com/sensors/sensors-18-00183/article_deploy/html/images/sensors-18-00183-g001.png
[5]. Lobov, S., Krilova, N., Kastalskiy, I., Kazantsev, V., & Makarov, V. A. (2018). Latent factors limiting the performance of
sEMG-interfaces. Sensors, 18(4), 1122.
[6]. Alejandro Mora Rubio, J. A. A. G., Reinel Tabares-Soto ORCID logo, Simón Orozco-Arias, Cristian Felipe Jiménez Varón, Jorge
Iván Padilla Buriticá (2020). Identification of Hand Movements from Electromyographic Signals Using Machine
Learning. doi: doi: 10.20944/preprints202002.0443.v1
[7]. Benalcázar, M. E., Jaramillo, A. G., Zea, A., Páez, A., & Andaluz, V. H. (2017). Hand gesture recognition using
machine learning and the Myo armband. Paper presented at the 2017 25th European Signal Processing Conference
(EUSIPCO).
References

40. Continued…
[8]. https://archive.ics.uci.edu/ml/datasets/EMG+data+for+gestures
[9]. Chawathe, S. S. (2019). Hand Gestures from Low-Cost Surface-Electromyographs. IEEE National Aerospace and
Electronics Conference (NAECON).
[10]. Benalcázar, M. E., Motoche, C., Zea, J. A., Jaramillo, A. G., Anchundia, C. E., Zambrano, P., . . . Pérez, M. (2017). Real-
time hand gesture recognition using the Myo armband and muscle activity detection. Paper presented at the 2017
IEEE Second Ecuador Technical Chapters Meeting (ETCM).