TinyML Approach for Pre-fall Motion Pattern Detection in Older Adults

  • Jefferson Sarmiento-Rojas
  • , Angela Maria Torres-Lara
  • , Pedro Antonio Aya Parra
  • , Jonnier Sebastián Jaramillo-Isaza
  • , Oscar Julian Perdomo

Research output: Chapter in Book/InformConference contribution

Abstract

This work presents a TinyML-based approach for classifying daily motion patterns and detecting pre-fall activity in older adults using inertial sensor data. A multilayer perceptron neural network was trained on six input signals (3-axis accelerometer and gyroscope) from the SisFall dataset. The model was optimized for edge deployment using quantization and pruning, achieving a global accuracy of 85.5% across 15 activity classes. The final model was converted to C for integration into microcontrollers. Results show high performance with minimal latency, enabling real-time fall prevention strategies in embedded health monitoring systems.

Original languageEnglish (US)
Title of host publicationApplied Computer Sciences in Engineering - 12th Workshop on Engineering Applications, WEA 2025, Proceedings
EditorsJuan Carlos Figueroa-García, Elvis Eduardo Gaona-García, Jesús Alfonso López-Sotelo, John Freddy Moreno-Trujillo
PublisherSpringer Science and Business Media Deutschland GmbH
Pages156-166
Number of pages11
ISBN (Print)9783032082022
DOIs
StatePublished - 2026
Event12th Workshop on Engineering Applications, WEA 2025 - Cali, Colombia
Duration: Oct 29 2025Oct 31 2025

Publication series

NameCommunications in Computer and Information Science
Volume2701 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937

Conference

Conference12th Workshop on Engineering Applications, WEA 2025
Country/TerritoryColombia
CityCali
Period10/29/2510/31/25

All Science Journal Classification (ASJC) codes

  • General Computer Science
  • General Mathematics

Fingerprint

Dive into the research topics of 'TinyML Approach for Pre-fall Motion Pattern Detection in Older Adults'. Together they form a unique fingerprint.

Cite this