INCREASING THE EFFICIENCY OF EMG SIGNALS BY USING MACHINE LEARNING ALGORITHMS Emre Parlak 1 , Çağdaş Özer 2 , Mustafa Takaoğlu 3 1 İstanbul Aydın University, emreparlak@aydin.edu.tr 2 İstanbul Aydın University, cagdasozer@aydin.edu.tr 3 İstanbul Aydın University, mustafatakaoglu@aydin.edu.tr ABSTRACT In our study, long-term performance signals was measured; to minimize the changes in the characteristics of the signals due to long term performance of amputated arm patients, it is aimed to improve the read signals by using machine learning algorithms. In our study, the data obtained from the measurements we made through the Armband device of the right arms of seven people were used. While the data were obtained, the hand was turned into a fist, and this movement continued until fatigue occurred in the muscle. Naive Bayes, Generalized Linear Model, Logistic Regression, Fast Large Margin, Deep Learning, Decision Tree, Random Forest, and Gradient Boosted Trees algorithms are used to process signals, and 16796 models are created. Data were analyzed based on Accuracy, Classification Error, Area Under Curve, Precision, Recall, F Measure, Sensitivity, Specificity. The algorithms that yield the best results were determined in each variable, and the results were shared. Keywords: EMG Signals, Machine Learning Algorithms, Signal Processing, Signal Efficiency, MYO Armband. 1. INTRODUCTION Many prosthetic arms have been developed for amputees since the 1970s [1]. In the process from 1970 to the present, studies in this field have continued increasingly. The loss of a limb, whether through congenital amputation, disease, or injury, is a traumatic experience. Many approaches have been proposed for the design of powered prosthetic devices to replace lost limb function since the concept of an EMG-controlled prosthetic hand was proposed by Wiener. Electromyogram (EMG) signals have often been used as control signals for prosthetic systems, such as the Utah artificial arm and the Boston arm by MIT [4]. In our study, we worked with EMG signals. The use of electromyographic (EMG) signals from skeletal muscle advantages of being both convenient and natural [2]. That means EMG signals are directly correlated with the contraction and relaxation of muscle fibers [3]. We’re using MYO Armband to detect EMG signals. Our purpose is to maximize signal detection when long term muscle movements happened. When muscles are tired due to long term usage, characteristic of the signal has been changing. With the help of machine learning algorithms, we tried to increase the success of the muscle signal process. To do that, we use MYO Armband on seven volunteers and get right arm muscle data. After collecting tired muscle data, we use them on eight different machine learning algorithms. Results have been shared on the Results and Discussion part. The importance of our study is to share the results of different machine learning algorithms that worked with our dataset. 2. MATERIALS AND METHODS In this study, muscle signal measurements were performed with MYO Armband produced by Thalmic Labs. MYO Armband is a wristband with 8 EMG electrodes, 3-axis accelerometer, 3-axis gyroscope and 3-axis magnetic force measurement. This wrist strap can also show the orientation of the arm in 3-dimensional space. The orientation data from the wristband is transmitted to the computer via wireless communication (Bluetooth). After the data is processed with the software prepared in Python programming language, it is sent to the industrial robot in real-time via TCP / IP communication [5].