Enhancing accuracy in fall detection and prediction for elderly individuals using ensemble wavelet neural network and maximal overlap discrete wavelet transform

Falls are the second-largest risk factor for the health of the elderly. Various researchers have proposed approaches for monitoring health and falls using only the Internet of Things (IoT) and sensors. However, relying on a single sensor limits accuracy in fall risk prediction. Data from multiple sensors can be used to classify human posture through learning-based algorithms stemming from machine learning. Recently, posture-detecting sensors have become increasingly popular, with InvenSense's Inertial Measurement Unit 9250 (IMU 9250) being a notable example for detecting elderly motion and transmitting data to a computer. This paper proposes the Ensemble Wavelet Network Framework (EWNNET) for fall prediction and detection using the Maximal Overlap Discrete Wavelet Transform (MODWT). The inertial measurement unit (IMU) dataset, containing real-time posture data from seven sensors, is split into 80 % for training and 20 % for testing. The EWNNET approach is compared with other machine-learning methods for fall detection, showing improved accuracy, sensitivity, precision, F-score, and specificity, with a 96.7 % classification success rate.