Physical Activity Recognition Based on a Parallel Approach for an Ensemble of Machine Learning and Deep Learning Classifiers.

Mariem Abid,Hugo Watel,Amar Mitiche,Neila Mezghani,Amel Benazza-Benyahia,Safouene Chemcki,Youssef Ouakrim,Amal Khabou

doi:10.3390/s21144713

Abstract

Human activity recognition (HAR) by wearable sensor devices embedded in the Internet of things (IOT) can play a significant role in remote health monitoring and emergency notification to provide healthcare of higher standards. The purpose of this study is to investigate a human activity recognition method of accrued decision accuracy and speed of execution to be applicable in healthcare. This method classifies wearable sensor acceleration time series data of human movement using an efficient classifier combination of feature engineering-based and feature learning-based data representation. Leave-one-subject-out cross-validation of the method with data acquired from 44 subjects wearing a single waist-worn accelerometer on a smart textile, and engaged in a variety of 10 activities, yielded an average recognition rate of 90%, performing significantly better than individual classifiers. The method easily accommodates functional and computational parallelization to bring execution time significantly down.

Highlights

Miniaturization of complex electrical devices at continually lower cost has brought about the development of a variety of wearable sensors and their embedding in healthcarededicated Internet of things (IoT)
If we examine the recognition performance for each activity individually, walking up the stairs is confused with walking (0.05%), walking down the stairs is confused with running (0.03%), and walking is confused with walking up the stairs (0.002%) and down the stairs (0.001%)
We compare the ensemble approach that we proposed to a selection of classifier algorithms including an ensemble approach, referred to as ensemble (DT-LR-MLP), that combines J48 decision tree, Multilayer Perceptrons (MLPs) and Logistic Regression [38] along with Random Forest [60] and Adaboost [61] using the default parameters available in the sklearn library

Summary

Introduction

Miniaturization of complex electrical devices at continually lower cost has brought about the development of a variety of wearable sensors and their embedding in healthcarededicated Internet of things (IoT). The broad purpose of a healthcare IoT, sometimes called Internet of medical things, abbreviated as IoMT, is to provide a network of embedded systems to acquire, communicate, and analyze data for remote medical practice of accrued quality. [2] provided a qualitative synthesis of studies using wearable body sensors for health monitoring, pointing out a number of shortcomings in prior research with respect to both sample size and participant demographics. Such systems aim at developing methods for automatically recognizing human physical activities by analyzing data gathered by sensors in wearable devices. The basic problem is to assign a time series segment of sensor data to a corresponding activity during that time segment [3]

Objectives

Methods

Results

Conclusion