A Fall Detection Algorithm for Thigh mounted Smartphones using Random Forest and Feature Selection Techniques

Falls are sudden incidents that cause the body to go down onto the ground or toward the ground without intention or by accident. Falls can result in physical injuries and mental distress, especially in the elderly. Such results can cause serious disabilities and in the worst cases even death. Conseq...

Full description

Saved in:
Bibliographic Details
Published in:2022 17th International Conference on Emerging Technologies (ICET) pp. 48 - 53
Main Authors: Maryam, Rida, Shahzad, Ahsan, Bashir, Mariam, Caceres-Najarro, Lismer Andres, Kim, Minseok
Format: Conference Proceeding
Language:English
Published: IEEE 29.11.2022
Subjects:
Classification algorithms
Entropy
Fall detection
Fall detection algorithm
Feature extraction
features selection
random forest
Simulation
smartphones
Thigh
Transfer learning
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Falls are sudden incidents that cause the body to go down onto the ground or toward the ground without intention or by accident. Falls can result in physical injuries and mental distress, especially in the elderly. Such results can cause serious disabilities and in the worst cases even death. Consequently, falls among the elderly are a widespread concern. That is one of the reasons why research on fall detection and prevention systems has gained considerable attention. Recently, smartphone (SP) based fall detection systems have attracted a lot of research efforts. Although there are several works using SPs for fall detection systems, most of them need to use additional external devices in addition to the SP and/or require the SPs to be positioned with a fixed orientation limiting their applicability. In daily life, people, most of the time, carry the SP in thigh locations such as the pockets of the pants. Therefore, this study focuses on the development of a reliable and highly accurate fall detection algorithm using thigh position SPs signals. The proposed algorithm employs a random forest classifier that incorporates feature selection techniques for an adequate selection of the best features. The proposed algorithm is trained and tested on the SP dataset used in a recent algorithm called FallDroid. Our simulation results show that the proposed approach achieves excellent fall detection scores outperforming the FallDroid.
DOI:10.1109/ICET56601.2022.10004626