HUMAN ACTIVITY RECOGNITION THROUGH ENSEMBLE LEARNING OF MULTIPLE CONVOLUTIONAL NEURAL NETWORKS

Authors

  • B.Nagalakshmi Assistant Professor, Department of IT, Lords Institute of Engineering and Technology, Hyderabad Author
  • Mohammad Abdullah, Abdul Imran Khan, Syed Ilahdad Mouzzam B.E. Student, Department of IT, Lords Institute of Engineering and Technology, Hyderabad Author

Abstract

Human Activity Recognition (HAR) focuses on identifying physical human activities through the analysis
of sensor data, such as one-dimensional time series data. Traditionally, this task has relied on hand-crafted features
to build machine learning models, a process that demands substantial domain expertise and feature engineering.
However, with advancements in deep neural networks, models can now autonomously learn features from raw
sensor data, resulting in enhanced classification performance. In this paper, we propose an innovative method for
human activity recognition using an ensemble approach that combines multiple convolutional neural network
(CNN) models.
We trained three distinct CNN models on a publicly available dataset and created several ensembles of these
models. Notably, the ensemble of the first two models achieved an accuracy of 94%, surpassing existing methods in
the literature. Index Terms—Human activity recognition, ensemble learning, deep learning, convolutional neural
networks.[

Downloads

Published

2024-08-28

Issue

Vol. 14 No. 3 (2024): July-September

Section

Articles

How to Cite

HUMAN ACTIVITY RECOGNITION THROUGH ENSEMBLE LEARNING OF MULTIPLE CONVOLUTIONAL NEURAL NETWORKS. (2024). International Journal of Engineering and Science Research, 14(3), 496-506. https://www.ijesr.org/index.php/ijesr/article/view/945