Paperdoi:10.1002/aisy.70653

Deployable Stimming Classification From Wrist-Worn 9 Degrees of Freedom Inertial Sensing: Accuracy, Memory, and Latency Trade-Offs

Venue listedDated Nov 15, 2026Added Nov 17, 2026

  • MA
  • NL
  • ER
  • KH
  • SR
  • AE
  • KS

Authors: Montserrat Ramirez-De Angel, Neil Joshua Limbaga, Eckaard le Roux, Kareem Hamdan, Samanta Castro Rodriguez, Ahmed M. Eltawil, Khaled N. Salama

Advanced Intelligent Systems (Wiley), published 2026-11-15

TL;DR

Wrist 9DoF node classifying four stimming behaviours vs two control activities, fully subject-independent. Feature-based classifiers all exceed macro-F1 0.97, and the accelerometer alone separates stimming from non-stimming at F1 above 0.998, so accuracy is not the bottleneck. On the target microcontroller, inference ranged from 3.7 to 631 ms and the ranking changed from the workstation one: memory and latency, measured on device, decide model choice.

Abstract

Objective monitoring of repetitive motor behaviors, or stimming, could support quantitative assessment in autism research and assistive technology, but wearable systems reported to date are typically validated on very few participants and benchmarked on desktop hardware, leaving their actual deployability unresolved. Here, a compact wrist‐worn nine‐degree‐of‐freedom (9DoF) inertial sensing node is presented together with a deployment‐oriented rule for selecting a wearable classifier. Recordings from participants performing four stimming behaviors and two nonstimming control activities are used to compare three feature‐based classifiers and four deep‐learning architectures across four sensing configurations under fully subject‐independent validation. Accuracy is found not to be the limiting factor: across the feature‐based classifiers, macro‐ F 1 exceeds 0.97 for every model and sensing configuration, and stimming is separated from nonstimming activity with an F 1 above 0.998 using the accelerometer alone. Deployment cost, by contrast, varies by more than an order of magnitude, and porting the networks to the target microcontroller reveals on‐device inference spanning 3.7–631 ms, modifying the ranking obtained on a workstation. Model selection for wearable stimming monitoring is therefore governed by memory and latency rather than by accuracy, and on‐device measurement is shown to be a prerequisite rather than a refinement.

Verbatim from the publisher's DOI record (opens in a new tab) · Publisher licence: http://creativecommons.org/licenses/by/4.0/

Results on ChatPPG leaderboards

All leaderboards

No results from this paper on ChatPPG leaderboards yet. Browse the leaderboards.

Tasks & signals

Tasks

Activity recognition

Signals

Keywords

HarOn DeviceMicrocontrollerTinymlLatencyMemoryModel SelectionSubject Independent

How it works

How Method works: Setup → Method → TargetSetup: Signals (accelerometer, gyroscope, magnetometer). Method: Method. Target: Activity recognition.SetupSignals: accelerometer, gyroscope, magnetometerxyzSignalsaccelerometer, gyroscope…MethodMethodTargetActivity recognitionActivity recognition

Cite this paper

Cite this paper with the fields recorded here. Check it against the publisher before submitting.

Angel, M. R.-D., Limbaga, N. J., le Roux, E., Hamdan, K., Rodriguez, S. C., Eltawil, A. M., & Salama, K. N. (2026). Deployable Stimming Classification From Wrist-Worn 9 Degrees of Freedom Inertial Sensing: Accuracy, Memory, and Latency Trade-Offs. Advanced Intelligent Systems. https://doi.org/10.1002/aisy.70653
Angel, Montserrat Ramirez-De, Neil Joshua Limbaga, Eckaard le Roux, Kareem Hamdan, Samanta Castro Rodriguez, Ahmed M. Eltawil, and Khaled N. Salama. 2026. "Deployable Stimming Classification From Wrist-Worn 9 Degrees of Freedom Inertial Sensing: Accuracy, Memory, and Latency Trade-Offs." Advanced Intelligent Systems. https://doi.org/10.1002/aisy.70653.
Angel, M. R.-D., Limbaga, N. J., le Roux, E., Hamdan, K., Rodriguez, S. C., Eltawil, A. M. and Salama, K. N. (2026) 'Deployable Stimming Classification From Wrist-Worn 9 Degrees of Freedom Inertial Sensing: Accuracy, Memory, and Latency Trade-Offs', Advanced Intelligent Systems. doi: 10.1002/aisy.70653.
Angel, Montserrat Ramirez-De, et al. "Deployable Stimming Classification From Wrist-Worn 9 Degrees of Freedom Inertial Sensing: Accuracy, Memory, and Latency Trade-Offs." Advanced Intelligent Systems, 2026. https://doi.org/10.1002/aisy.70653.
M. R.-D. Angel et al., "Deployable Stimming Classification From Wrist-Worn 9 Degrees of Freedom Inertial Sensing: Accuracy, Memory, and Latency Trade-Offs," Advanced Intelligent Systems, 2026, doi: 10.1002/aisy.70653.
@article{angel2026deployable,
  title         = {{Deployable Stimming Classification From Wrist-Worn 9 Degrees of Freedom Inertial Sensing: Accuracy, Memory, and Latency Trade-Offs}},
  author        = {Angel, Montserrat Ramirez-De and Limbaga, Neil Joshua and le Roux, Eckaard and Hamdan, Kareem and Rodriguez, Samanta Castro and Eltawil, Ahmed M. and Salama, Khaled N.},
  year          = {2026},
  journal       = {Advanced Intelligent Systems},
  publisher     = {Wiley},
  doi           = {10.1002/aisy.70653}
}

Added to the ChatPPG research hub