A Comparative Study of Machine Learning and Deep Neural Architectures for EMG-Based Prosthetic Hand Gesture Recognition

Abstract

Surface electromyography (sEMG) has emerged as a widely adopted non-invasive interface for prosthetic hand control due to its ability to capture muscle activation patterns associated with user intent. However, the inherent non-stationary nature of EMG signals, together with noise, motion artifacts and physiological variability, continues to present challenges for reliable gesture recognition. This study presents a comparative evaluation of classical machine learning and deep learning approaches for EMG-based prosthetic hand gesture recognition.EMG signals were acquired from forearm muscles using a BIOPAC Systems Inc. MP150 data acquisition system operating in a single-channel configuration with a sampling frequency of 400 Hz. The recorded signals were pre-processed using band-pass and notch filtering, followed by normalization and sliding-window segmentation. A unified experimental framework was established to ensure consistent evaluation across all investigated models. Classical machine learning classifiers, including Support Vector Machine (SVM) and Random Forest (RF), were compared with deep learning architectures comprising a Convolutional Neural Network (CNN) and a Temporal Convolutional Network (TCN). Experimental results demonstrated successful classification of hand gestures across all evaluated models. The Random Forest and SVM classifiers achieved accuracies of 90.2% and 88.4%, while the CNN and TCN achieved accuracies of 94.1% and 95.8% respectively under the experimental conditions considered in this study. Among the investigated approaches, the TCN produced the highest overall performance, indicating the benefit of temporal modeling for EMG signal interpretation. The results further highlight the trade-off between classification accuracy and computational complexity, with classical machine learning methods offering lower implementation overhead and deep learning models providing superior recognition performance.The findings suggest that temporal deep learning architectures represent a promising direction for EMG-based prosthetic control systems. Future work will focus on lightweight deployment strategies, multimodal sensing integration and adaptive learning techniques to improve robustness in practical assistive applications

Introduction

This study explores the use of surface electromyography (sEMG) signals for controlling prosthetic hands through accurate hand gesture recognition. Loss of upper-limb function significantly affects daily life, and EMG-based prosthetic systems provide a non-invasive way to capture muscle activity and translate user intent into prosthetic movements. However, EMG signals are noisy, non-stationary, and influenced by factors such as electrode placement, muscle fatigue, and physiological variations.

To improve gesture recognition, the study compares classical machine learning models (Support Vector Machine and Random Forest) with deep learning architectures (Convolutional Neural Network and Temporal Convolutional Network). EMG signals were collected from forearm muscles using a BIOPAC MP150 system at a sampling rate of 400 Hz. Signals were preprocessed using band-pass filtering (20–180 Hz), notch filtering (50 Hz), normalization, and segmented using a 200-sample sliding window with 50% overlap. For traditional machine learning models, features such as Mean Absolute Value (MAV), Root Mean Square (RMS), Waveform Length (WL), Zero Crossing (ZC), and Variance (VAR) were extracted, while CNN and TCN models used raw segmented EMG signals directly.

The dataset consisted of approximately 43 minutes of recordings, producing 2,800 labeled gesture segments across four hand gesture classes. An 80:20 train-test split was used, and model performance was evaluated using accuracy, precision, recall, and F1-score.

Results showed that deep learning models outperformed classical approaches. The Temporal Convolutional Network (TCN) achieved the highest accuracy (95.8%), followed by CNN (94.1%), Random Forest (90.2%), and SVM (88.4%). TCN performed best because it effectively captured temporal dependencies in EMG signals using dilated convolutions while maintaining computational efficiency.

The study concludes that deep learning, particularly TCN, offers superior performance for EMG-based hand gesture recognition and prosthetic control. It also highlights the importance of balancing classification accuracy, robustness, and computational requirements for real-time embedded prosthetic systems. The proposed evaluation framework provides insights into selecting suitable models for practical prosthetic applications and future deployment on resource-constrained devices.

Conclusion

This study presented a comparative evaluation of classical machine learning approaches and deep neural architectures for EMG-based hand gesture recognition. A unified processing framework incorporating signal acquisition, preprocessing, segmentation and classification was employed to ensure a consistent evaluation of the investigated models. Handcrafted time-domain descriptors were utilized for the classical machine learning classifiers, whereas the deep learning architectures operated directly on segmented EMG signals. Experimental results obtained from forearm EMG recordings demonstrated that all evaluated models were capable of successfully classifying hand gestures. Among the investigated approaches, the Temporal Convolutional Network achieved the highest classification accuracy, followed closely by the Convolutional Neural Network. The observed performance improvements suggest that temporal modeling can provide additional discriminative capability for EMG-based gesture recognition by capturing sequential characteristics present within muscle activation patterns.The classical machine learning models achieved competitive performance while maintaining lower computational complexity, highlighting their continued relevance for resource-constrained applications. In contrast, the deep learning models achieved superior classification accuracy at the cost of increased computational requirements. These findings emphasize the trade-off between recognition performance and deployment complexity that must be considered when designing practical prosthetic control systems. The results presented in this work were obtained under controlled experimental conditions and demonstrate the feasibility of both classical and deep learning approaches for EMG-based gesture classification. While the Temporal Convolutional Network produced the strongest overall performance within the evaluated experimental setup, further investigation involving larger and more diverse datasets would be valuable for assessing robustness across broader operating conditions. In conclusion, temporal deep learning architectures represent a promising direction for EMG-driven prosthetic control. Future developments incorporating model optimization, adaptive learning strategies and multimodal sensing may further enhance the reliability, efficiency and practical usability of intelligent prosthetic systems.

References

[1] Oskoei, M. A., & Hu, H. (2007). Myoelectric control systems—A survey. Biomedical Signal Processing and Control, 2(4), 275–294. [2] Farina, D., Jiang, N., Rehbaum, H., et al. (2014). The extraction of neural information from the surface EMG for the control of upper-limb prostheses: Emerging avenues and challenges. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 22(4), 797–809. [3] Hudgins, B., Parker, P., & Scott, R. N. (2002). A new strategy for multifunction myoelectric control. IEEE Transactions on Biomedical Engineering, 40(1), 82–94. [4] Jiang, N., Dosen, S., Müller, K., & Farina, D. (2012). Myoelectric control of artificial limbs: Is there a need to change focus? IEEE Signal Processing Magazine, 29(5), 150–152. [5] Farina, D., & Merletti, R. (2000). Comparison of algorithms for estimation of EMG variables during voluntary isometric contractions. Journal of Electromyography and Kinesiology, 10(5), 337–349. [6] Ison, M., & Artemiadis, P. (2014). The role of muscle synergies in myoelectric control: trends and challenges for simultaneous multifunction control. Journal of neural engineering, 11(5), 051001. [7] Zecca, M., Micera, S., Carrozza, M. C., & Dario, P. (2002). Control of multifunctional prosthetic hands by processing the electromyographic signal. Critical Reviews in Biomedical Engineering, 30(4–6), 459–485. [8] Englehart, K., & Hudgins, B. (2003). A robust, real-time control scheme for multifunction myoelectric control. IEEE Transactions on Biomedical Engineering, 50(7), 848–854. [9] Phinyomark, A., Phukpattaranont, P., & Limsakul, C. (2012). Feature reduction and selection for EMG signal classification. Expert Systems with Applications, 39(8), 7420–7431. [10] Li, G., Schultz, A., & Kuiken, T. (2010). Quantifying pattern recognition-based myoelectric control of multifunctional transradial prostheses. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 18(2), 185–192. [11] LeCun, Y., Bengio, Y., & Hinton, G. (2015). Deep learning. Nature, 521(7553), 436–444. [12] Hashi, A. O., Hashim, S. Z. M., & Asamah, A. B. (2024). A systematic review of hand gesture recognition: An update from 2018 to 2024. IEEE Access, 12, 143599-143626. [13] Wei, W., Dai, Q., Wong, Y., Hu, Y., Kankanhalli, M., & Geng, W. (2019). Surface-electromyography-based gesture recognition by multi-view deep learning. IEEE Transactions on Biomedical Engineering, 66(10), 2964–2973. [14] Zhang, Y., Yang, F., Fan, Q., Yang, A., & Li, X. (2022). Research on sEMG-based gesture recognition by dual-view deep learning. IEEE Access, 10, 32928–32937. [15] Atzori, M., Gijsberts, A., Castellini, C., Caputo, B., Hager, A. G. M., Elsig, S., et al. (2014). Electromyography data for non-invasive naturally-controlled robotic hand prostheses. Scientific Data, 1(1), 140053. [16] Ameri, A., Scheme, E., & Englehart, K. (2019). A deep transfer learning approach to reducing the effect of electrode shift in EMG pattern recognition-based control. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 27(2), 370–379. [17] Phinyomark, A., Scheme, E., & Englehart, K. (2018). Feature extraction and selection for myoelectric control based on wearable EMG sensors. Sensors, 18(5), 1615. [18] Boostani, R., & Moradi, M. H. (2003). Evaluation of the forearm EMG signal features for the control of a prosthetic hand. Physiological Measurement, 24(2), 309–319. [19] Bai, S., Kolter, J. Z., & Koltun, V. (2018). An empirical evaluation of generic convolutional and recurrent networks for sequence modeling.

Copyright

Copyright © 2026 Abhishek ., A. J. Singh. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.