A methodology to rank importance of frequencies and channels in electromyography data with Decision Tree classifiers

Albert A. Nasybullin, Nursultan Abdullaev, Maksim A. Baranov, Vitaly A. Mahonin · 2026

Read A methodology to rank importance of frequencies and channels in electromyography data with Decision Tree classifiers by Albert A. Nasybullin, Nursultan Abdullaev, Maksim A. Baranov, Vitaly A. Mahonin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-15353.txt.

Introduction

Nasybullin and four colleagues offer a method for ranking which frequencies and channels in electromyography signals matter most for judging recovery. Using vastus lateralis signals from squats at varied rest intervals, single decision trees rank features by power-spectrum importance with full transparency. A small set of top features already gives reliable rest-interval classification, pointing toward compact models for medical and sports use.

Catalog record

CREATORAlbert A. Nasybullin, Nursultan Abdullaev, Maksim A. Baranov, Vitaly A. Mahonin
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE21452
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.15353

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

A methodology to rank importance of frequencies and channels in electromyography data with Decision Tree classifiers Thanks: [email protected] Albert A. Nasybullin, Nursultan Abdullaev, Maksim A. Baranov, Vitaly A. Mahonin Abstract This study presents a methodology for identifying the most informative frequencies and channels in electromyography (EMG) data to evaluate muscle recovery using Decision Tree classifiers. EMG signals, recorded from the vastus lateralis muscle during squat exercises, were analyzed across varying rest intervals to assess optimal recovery periods. By employing single Decision Tree classifiers, the study enhances interpretability, offering insights into feature importance - essential for applications in medical and sports settings where transparency is critical. The experimental protocol utilized a grid search for hyperparameter tuning and cross-validation to address class imbalance, ultimately achieving a reliable classification of rest intervals based on power spectral density features. The results indicate that a limited subset of highly informative features provides sufficient accuracy, suggesting that streamlined, interpretable models are effective for the evaluation of muscle recovery. This approach can guide future research in developing compact, robust models adapted to EMG-based diagnostics. Keywords: Electromyography (EMG), Signal Classification, Decision Tree Classifier, Feature Importance, Biomedical Signal Processing, Muscle Recovery 1 Introduction Electromyography (EMG) is widely used in biomedical and sports sciences to analyze muscle activity, assess neuromuscular function, and monitor recovery following exertion. EMG signals, reflecting muscle electrical activity, are complex and often noisy, making the identification of informative features essential for effective analysis. Machine learning algorithms, particularly decision trees, have proven effective in EMG classification due to their capacity to capture non-linear relationships and handle complex data. Recent studies on EMG classification frequently employ tree-based ensemble methods for their robustness and improved accuracy with high-dimensional, noisy data Seyidbayli et al. 2020 ; Ramírez-Pérez et al. 2024 ; Abdullah et al. 2017 . However, ensemble models typically lack interpretability — a critical factor in medical and sports applications where transparency is paramount, as well as for developers seeking insights into model behavior and feature contributions. In this study, we use single Decision Tree classifiers to identify the most significant features in EMG signals. This approach enables interpretable assessment of feature importance, revealing which signal characteristics are most relevant for further analysis and model development. We aim to identify a set of high-performing features to guide future researchers in designing efficient EMG classification models. Pinpointing key frequencies and channels that impact classification accuracy provides a foundation for subsequent studies to develop compact, effective models tailored to muscle recovery and EMG-based diagnostics. 2 Data collection and data structure The initial sample comprised 14 volunteers, including 11 males and 3 females (78.6% males and 21.4% females), aged between 18 and 43 years (mean age = 24.86 years, standard deviation = 7.86). All participants were right-handed, non-smokers, and engaged in regular physical activity. To control for confounding variables, participants were instructed to abstain from alcohol consumption and physical exercise for 48 hours prior to the experiment. They reported being well-rested and fully recovered from their most recent physical activity sessions. Each participant indicated that their caffeine consumption levels were consistent with their typical daily intake. None reported any diagnosed musculoskeletal or central nervous system disorders, nor were they under prescribed medication. During the exploratory data analysis, three male participants were excluded due to hardware malfunctions and poor data quality. An additional male participant was excluded because of a prior spinal injury. The final sample thus consisted of 10 participants: 7 males and 3 females (70% males and 30% females), aged between 18 and 43 years (mean age = 24.5 years, standard deviation = 7.65). Prior to the commencement of the experiment, all participants were informed about the study’s objectives, methodologies, and the potential for minor discomforts. 2.1 Experimental Design Figure 1: The figure illustrates an experimental protocol for recording muscle activity (EMG) during repeated squat and stand actions. The sequence consists of two main actions, each held for 3 seconds: squatting to a 90 ∘ knee angle with the command ”close” and standing upright with the command ”open.” This cycle repeats three times within each block. After completing the squat-stand cycle, participants enter a resting period lasting 1, 5, or 10 minutes, depending on the block. The objective of this experiment is to investigate optimal rest intervals that allow muscle recovery following physical exertion Valeriano et al. 2023 ; Mcmahon et al. 2024 . For the exercise, we selected squats, where participants are required to hold each position (squat or standing) for a duration of three seconds (Fig. 1 ). Figure 2: The experimental design consists of three blocks of varying lengths (1, 5, or 10 minutes), each separated by a 10-minute rest interval. Each block includes three sample periods interspersed with rest intervals equal to the block length (K). Following the samples, there is a baseline period, succeeded by three sequential activities (Seq 0, Seq 1, and Seq 2). The sequence includes two specific events: a ”squat” event and a ”stand” event. Our experimental setup (Fig. 2 ) is divided into three blocks, each exploring different rest intervals: k\in\{1,5,10\} minutes Mcmahon et al. 2024 . Between each block, there is a standard 10-minute break, during which the participant rests in a seated position to allow for muscle recovery. Within each block, participants complete three samples, with a rest interval of k minutes between each other. Each sample consists of a baseline measurement followed by three sequences of actions. In each sequence, there are two main events: a three-second squat hold and a three-second standing hold (Fig. 1 ). 2.2 Data Acquisition and Preprocessing for EMG Analysis Figure 3: The image shows the electrode placement for EMG recording on the vastus lateralis muscle. Three EMG channels are set up along the thigh, each using a green electrode as the positive (+) terminal, a yellow electrode as the negative (-) terminal, and a red electrode as the reference (ref). The left panel displays the actual electrode placement on a participant’s leg, while the right panel illustrates the anatomical positioning on the vastus lateralis muscle, highlighting the alignment of each EMG channel along the muscle. Data was recorded using electromyography (EMG) Merletti and Muceli 2019 . For this purpose, we developed a custom three-channel setup to monitor and analyze the vastus lateralis muscle (Fig. 3 ). 2.2.1 System Overview The foundation of our system is built on the STM32F401 microcontroller Darmawan et al. 2022 , supported by external electromyography (EMG) signal amplifiers. The microcontroller’s integrated analog-to-digital converter (ADC) captures signals, and data transmission occurs at 1 kHz Komorowski et al. 2024 . Direct Memory Access (DMA) is used for efficient data handling, with the data organized into packets and sent via a virtual serial port to the connected computer for further analysis. 2.2.2 Data Preprocessing Upon obtaining the experimental sample data, we perform preprocessing on the EMG signals. The initial step involves baseline correction, where we calculate the average value from a 5-second segment of the recording. Following this, the signal undergoes bandpass filtering within the frequency range of 20 to 500 Hz Cavalcanti Garcia and Vieira 2011 ; Jamal 2012 . Additionally, we employ a notch filter to eliminate power line noise at 50 Hz Jamal 2012 ; Komorowski et al. 2024

Read the complete text