Detecting Unseen Falls from Wearable Devices using Channel-wise Ensemble of Autoencoders

TL;DR

This paper introduces an ensemble autoencoder approach for detecting unseen falls from wearable sensor data, addressing challenges in feature selection and threshold setting with novel error tightening methods.

Contribution

It proposes a channel-wise ensemble autoencoder framework trained on normal activities and introduces two methods for automatic error threshold adjustment to improve unseen fall detection.

Findings

Outperforms traditional autoencoders in unseen fall detection

Effective on two activity recognition datasets

Improves detection accuracy over standard one-class classifiers

Abstract

A fall is an abnormal activity that occurs rarely, so it is hard to collect real data for falls. It is, therefore, difficult to use supervised learning methods to automatically detect falls. Another challenge in using machine learning methods to automatically detect falls is the choice of engineered features. In this paper, we propose to use an ensemble of autoencoders to extract features from different channels of wearable sensor data trained only on normal activities. We show that the traditional approach of choosing a threshold as the maximum of the reconstruction error on the training normal data is not the right way to identify unseen falls. We propose two methods for automatic tightening of reconstruction error from only the normal activities for better identification of unseen falls. We present our results on two activity recognition datasets and show the efficacy of our proposed method against traditional autoencoder models and two standard one-class classification methods.