Design of wearable device for elderly fall detection using ensemble learning

Abstrak

Abstract This study presents an artificial intelligence (AI) approach for fall detection in elderly care using inertial measurement unit (IMU) data. IMU sensors were chosen for their high accuracy in capturing multi-axis linear acceleration and angular velocity, compact size, low power consumption, ease of integration into wearable devices, and ability to preserve user privacy compared to vision-based systems. The proposed method employs a stacking ensemble model combining Random Forest and Gradient Boosting as base learners with Logistic Regression as the meta-learner. Raw accelerometer and gyroscope signals (100 Hz) were segmented using a 3 s sliding window with 50% overlap, followed by the extraction of Hjorth descriptors and statistical features. Automatic labeling was performed using a mobility-based threshold. The trained model achieved an F1-score of 0.92 and an AUC of 0.969, outperforming individual base models. To validate real-world applicability, the Random Forest classifier was optimized and implemented in C for deployment on an embedded microcontroller (Arduino Nano RP2040 Connect) with Bluetooth Low Energy (BLE) capability for real-time fall alerts.

Identifier

• OpenAlex: https://openalex.org/W7138920957
• DOI: https://doi.org/10.1088/1742-6596/3188/1/012016