Fall detection using a Gaussian distribution of clustered knowledge, augmented radial basis neural-network, and multilayer perceptron

The rapidly increasing population of elderly people has posed a big challenge to research in fall prevention and detection. Substantial amounts of injuries, disabilities, traumas and deaths among elderly people due to falls have been reported worldwide. There is therefore a need for a reliable, simple, and affordable automatic fall detection system. This paper proposes a reliable fall detection algorithm using minimal information from a single waist worn wireless tri-axial accelerometer. The method proposed is to approach fall detection using digital signal processing and neural networks. This method includes the application of Discrete Wavelet Transform (DWT), Regrouping Particle Swarm Optimization (RegPSO), a proposed method called Gaussian Distribution of Clustered Knowledge (GCK), and an Ensemble of Classifiers using two different classifiers: Multilayer Perceptron Neural Network (MLP) and Augmented Radial Basis Neural Networks (ARBF). The proposed method has been tested on 8 healthy individuals in a home environment and yields promising result of up to 100% sensitivity on ingroup, 97.65% sensitivity on outgroup, and 99.56% specificity on Activities of Daily Living (ADL) data.