Implementing ensemble of deep learning model with optimization techniques for human activity recognition to assist...

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Human Activity Recognition (HAR) is vital for applications in healthcare, smart homes, and manufacturing.
• HAR systems analyze human behavior and movements to assist individuals, particularly those with disabilities.
• Existing HAR methods require robust models for accurate disability assistance.

Purpose of the Study:

• To propose a Binary Grey Wolf Optimization-driven Ensemble Deep Learning Model (BGWO-EDLMHAR) for enhanced Human Activity Recognition (HAR).
• To develop a robust HAR technique specifically for disability assistance applications.
• To improve the accuracy and efficiency of HAR systems through advanced optimization and deep learning.

Main Methods:

• Data preprocessing involved z-score normalization for data structuring.
• Feature selection was performed using the Binary Grey Wolf Optimization (BGWO) method.
• An ensemble of deep learning models, including Bidirectional Long Short-Term Memory, Variational Autoencoder, and Temporal Convolutional Network, was utilized for classification.
• Hyperparameter tuning was optimized using the Cetacean Optimization Algorithm.

Main Results:

• The BGWO-EDLMHAR technique demonstrated superior performance on the WISDM dataset.
• The proposed model achieved a high classification accuracy of 98.51%.
• Comparative analysis confirmed the superior accuracy of BGWO-EDLMHAR over existing HAR models.

Conclusions:

• The BGWO-EDLMHAR technique offers a robust and accurate solution for human activity recognition.
• This approach significantly enhances the potential for HAR in disability assistance.
• The integration of advanced optimization and ensemble deep learning models proves effective for HAR tasks.