Home // SENSORCOMM 2025, The Nineteenth International Conference on Sensor Technologies and Applications // View article

A Time based Sensor Data Analysis for Pre-Fall Prediction Using Machine and Deep Learning Approaches

Authors:
Nazia Pathan
Hongnian Yu
Michael Vassallo
Pelagia Koufaki

Keywords: Fall, Pre-Fall, Machine learning, Deep Learning.

Abstract:
Falls are a major cause of injury among older people, often leading to severe consequences, including death. To reduce this risk for both older and younger populations, Artificial Intelligence (AI) can play a critical role by predicting pre-fall states (conditions leading to a fall) and enabling timely intervention. Pre- fall prediction can be approached through various contexts, such as time-based, biological, and sensor data. This study focuses on predicting pre-falls through the time-based context by using the data from wearable sensors (accelerometer and gyroscope), while considering the time window feature of the dataset. The dataset used in this paper was collected using a MetaMotionR device and comprises two classes: “fall” and “no fall”. A sliding time window approach of 5 seconds and 10 seconds was applied to prepare the dataset for pre-fall prediction. Notably, this type of dataset has not previously been utilised for pre-fall prediction. A variety of machine learning and Deep Learning algorithms were tested on this dataset. The machine learning models included Decision Tree (DT), Support Vector Machine (SVM), and Logistic Regression (LR), and Deep Learning models included Deep Neural Networks (DNN), Convolutional Neural Networks (CNN), and Recurrent Neural Networks (RNN). Among machine learning algorithms, the DT demonstrated super performance, achieving accuracies of 95.99% and 95.75% for the 5-second and 10-second time windows, respectively. In the category of Deep Learning algorithms, Long Short Term Memory (LSTM) type of RNN models outperformed other approaches, with accuracies of 81.08% and 82.63% for the 5-sec and 10-sec windows, respectively.

Pages: 8 to 13

Copyright: Copyright (c) IARIA, 2025

Publication date: October 26, 2025

Published in: conference

ISSN: 2308-4405

ISBN: 978-1-68558-304-0

Location: Barcelona, Spain

Dates: from October 26, 2025 to October 30, 2025