A Hybrid Deep Learning-Based Approach for Human Activity Recognition Using Wearable Sensors

Deepak Sharma, Arup Roy, Sankar Prasad Bag, Pawan Kumar Singh, Youakim Badr

Research output: Chapter in Book/Report/Conference proceedingChapter


Human Activity Recognition (HAR) is a branch of computer science that uses raw time-series data information from embedded smartphone sensors and wearable devices to infer human actions. It has aroused considerable interest in various smart home contexts, particularly for constantly monitoring human behavior in an ecologically friendly atmosphere for elderly people and rehabilitation. Data collection, feature extraction from noise and distortion, feature selection, and pre-processing and categorization are among the operating components of a typical HAR system. Extraction of feature and selection strategies have recently been developed using cutting-edge approaches and traditional machine learning classifiers. The majority of the solutions, on the other hand, rely on simple feature extraction algorithms that are unable to detect complex behaviors. Deep learning techniques are often utilized in different HAR approaches to recover features and classification swiftly because of the introduction and development of vast computing resources. The vast majority of solutions, on the other hand, depend on simplistic feature extraction algorithms incapable of recognizing complicated behaviors. Due to advancements in high computational capabilities, deep learning algorithms are now often utilized in HAR methods to efficiently extract meaningful features which can successfully categorize sensor data. In this chapter, we present a hybrid deep learning-based classification model comprising of Convolutional Neural Network (CNN) and Long-Short Term Memory (LSTM), which is named CNN-LSTM. The proposed hybrid deep learning model has been tested over three benchmark HAR datasets: MHEALTH, OPPORTUNITY, and HARTH. On the aforementioned datasets, the proposed hybrid model obtained 99.07%, 95.2%, and 94.68% classification accuracies, respectively, which is quite impressive. The source code of the proposed work can be accessed by using the following link: https://github.com/DSharma05/Human-Activity-Recognition-using-hybrid-Deep-learning-approach.

Original languageEnglish (US)
Title of host publicationStudies in Big Data
PublisherSpringer Science and Business Media Deutschland GmbH
Number of pages29
StatePublished - 2023

Publication series

NameStudies in Big Data
ISSN (Print)2197-6503
ISSN (Electronic)2197-6511

All Science Journal Classification (ASJC) codes

  • Control and Systems Engineering
  • Engineering (miscellaneous)
  • Computer Science Applications
  • Artificial Intelligence

Cite this