Investigation of the segmentation window size for tasks of classifying the type of physical exercise based on data from the accelerometer and gyroscope of a smartphone

This article analyzes the effect of the size of the segmentation window on the quality of classification of the type of physical exercise based on data from the accelerometer and gyroscope of a smartphone. The article gives the concept and description of the HAR (Human Activity Recognition) task and its refinement for classifying specific types of physical exercises: squats, push-ups, jumps, abs, lunges. The review of existing data sets and approaches to solving problems of this class is carried out. The method of data collection for the experiment was chosen, and the attachment point of the device with sensors was determined. A tool (mobile application) has been developed to collect data from smartphone sensors such as accelerometer and gyroscope. Using the developed tool, a proprietary data set was collected under controlled conditions. The data obtained was processed based on general recommendations for the HAR class of tasks (data are reduced to a single frequency, noise-free, and segmented). Based on the obtained data sets, several models of both classical machine learning and deep neural networks with different parameters of the data segmentation window size were trained. As a result of the research, the best size of the data segmentation window was determined, as well as the classical machine learning and deep learning models that best performed the task.

1. Ann O.Ch., Theng L.B. Human Activity Recognition: A Review. In: 2014 IEEE International Conference on Control System, Computing and Engineering (ICCSCE 2014), 28–30 November 2014, Penang, Malaysia. IEEE; 2014. P. 389–393. https://doi.org/10.1109/ICCSCE.2014.7072750

2. Serpush F., Menhaj M.B., Masoumi B., Karasfi B. Wearable Sensor-Based Human Activity Recognition in the Smart Healthcare System. Computational Intelligence and Neuroscience. 2022;2022. https://doi.org/10.1155/2022/1391906

3. Kumar P., Suresh S. FLAAP: An Open Human Activity Recognition (HAR) Dataset for Learning and Finding the Associated Activity Patterns. Procedia Computer Science. 2022;212:64–73. https://doi.org/10.1016/j.procs.2022.10.208

4. Logacjov A., Bach K., Kongsvold A., Bårdstu H.B., Mork P.J. HARTH: A Human Activity Recognition Dataset for Machine Learning. Sensors. 2021;21(23). https://doi.org/10.3390/s21237853

5. Antonsson E.K., Mann R.W. The Frequency Content of Gait. Journal of Biomechanics. 1985;18(1):39–47. https://doi.org/10.1016/0021-9290(85)90043-0

6. Orlova Yu., Gorobtsov A., Sychev O., Rozaliev V., Zubkov A., Donsckaia A. Method for Determining the Dominant Type of Human Breathing Using Motion Capture and Machine Learning. Algorithms. 2023;16(5). https://doi.org/10.3390/a16050249

7. Dhaygude A.D., Varma R.A., Yerpude P., Swarnkar S.K., Jindal R.K., Rabbi F. Deep Learning Approaches for Feature Extraction in Big Data Analytics. In: 2023 10th IEEE Uttar Pradesh Section International Conference on Electrical, Electronics and Computer Engineering (UPCON), 01–03 December 2023, Gautam Buddha Nagar, India. IEEE; 2023. P. 964–969. https://doi.org/10.1109/UPCON59197.2023.10434607

8. Dargie W. Analysis of Time and Frequency Domain Features of Accelerometer Measurements. In: 2009 Proceedings of 18th International Conference on Computer Communications and Networks, 03–06 August 2009, San Francisco, CA, USA. IEEE; 2009. P. 1–6. https://doi.org/10.1109/ICCCN.2009.5235366

9. Avci A., Bosch S., Marin-Perianu M., Marin-Perianu R., Havinga P.J.M. Activity Recognition Using Inertial Sensing for Healthcare, Wellbeing and Sports Applications: A Survey. In: ARCS '10: 23th International Conference on Architecture of Computing Systems: Workshop Proceedings, 22–23 February 2010, Hannover, Germany. VDE Verlag; 2010. P. 167–176.

10. Powers D.M.W. What the F-Measure Doesn't Measure: Features, Flaws, Fallacies and Fixes. arXiv. URL: https://doi.org/10.48550/arXiv.1503.06410 [Accessed 20th March 2025].