Scholarlink

<p>The Human Activity Recognition (HAR) from the sensor component has developed the primary part for several real-world scenarios, such as healthcare and disease diagnosis. The conventional techniques have some drawbacks in terms of harmonizing accuracy and speed. Additionally, the relevant methods have not provided a solution for addressing the unfair data in dissimilar activities of HAR, even though it has the main concern for producing enhanced performance. This paper proposes the hybrid technique of Hierarchical Entropy-based GRU-CNN-AdaBoost Framework for performing the classification process of human activities. The proposed framework has the segmentation process with the sliding window concept. The dispersion entropy of several frequency units is produced by the feature vector group. Finally, the hybrid AdaBoost procedure is used to classify the human activities. The Hierarchical entropy computation has the processing of raw signals, frequency components, and time–frequency representations. The proposed model is used to extract the features for capturing temporal patterns in sensor data. The hybrid model produces a robust classification framework by learning spatial–temporal dependencies through a multi-class AdaBoost classifier. Comprehensive evaluation on the datasets of the KU-HAR dataset and USC-HAD dataset demonstrates that the proposed framework produces improved performance and maintains statistically significant enhancements (<i>p</i> &lt; 0.0001) over relevant techniques of GE-EnsemCNN-HAR, DeepConvLSTM, and ICGNet.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Accurate human activity recognition using hierarchical entropy-based GRU-CNN-AdaBoost framework

  • 期刊论文

  • 发表日期：

    2026年5月15日

查看全文
  • M. Bagya Lakshmi,
  • M. Nava Barathy

摘要

The Human Activity Recognition (HAR) from the sensor component has developed the primary part for several real-world scenarios, such as healthcare and disease diagnosis. The conventional techniques have some drawbacks in terms of harmonizing accuracy and speed. Additionally, the relevant methods have not provided a solution for addressing the unfair data in dissimilar activities of HAR, even though it has the main concern for producing enhanced performance. This paper proposes the hybrid technique of Hierarchical Entropy-based GRU-CNN-AdaBoost Framework for performing the classification process of human activities. The proposed framework has the segmentation process with the sliding window concept. The dispersion entropy of several frequency units is produced by the feature vector group. Finally, the hybrid AdaBoost procedure is used to classify the human activities. The Hierarchical entropy computation has the processing of raw signals, frequency components, and time–frequency representations. The proposed model is used to extract the features for capturing temporal patterns in sensor data. The hybrid model produces a robust classification framework by learning spatial–temporal dependencies through a multi-class AdaBoost classifier. Comprehensive evaluation on the datasets of the KU-HAR dataset and USC-HAD dataset demonstrates that the proposed framework produces improved performance and maintains statistically significant enhancements (p < 0.0001) over relevant techniques of GE-EnsemCNN-HAR, DeepConvLSTM, and ICGNet.