Scholarlink

<p>This study aims to improve human activity recognition (HAR) in smart home environments by proposing a dynamic segmentation method based on Spearman correlation. Traditional HAR approaches often rely on fixed-size windowing, which can struggle to capture the irregular and asynchronous nature of sensor activations. Our method addresses this limitation by dynamically adjusting segmentation boundaries in real-time, using both sensor correlation and temporal correlation to determine whether a new event belongs to an existing activity or starts a new one. Spearman correlation, a non-parametric rank-based metric, is particularly effective for identifying non-linear dependencies in sparse and noisy data, which are common in ambient sensor networks. We demonstrate the accuracy of our approach on the Aruba dataset from the CASAS project, a real-world smart home dataset. The suggested approach outperforms widely used baselines including Pearson correlation, sensor-event windowing, and decision trees by 11.5% to 17.0%, with an average F1-score of 0.854. These findings confirm that correlation-driven dynamic segmentation is a promising strategy for achieving accurate and efficient HAR in real-time smart home applications.</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

A novel dynamic segmentation approach based on spearman correlation for human activity recognition

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Khadija Essafi,
  • Laila Moussaid

摘要

This study aims to improve human activity recognition (HAR) in smart home environments by proposing a dynamic segmentation method based on Spearman correlation. Traditional HAR approaches often rely on fixed-size windowing, which can struggle to capture the irregular and asynchronous nature of sensor activations. Our method addresses this limitation by dynamically adjusting segmentation boundaries in real-time, using both sensor correlation and temporal correlation to determine whether a new event belongs to an existing activity or starts a new one. Spearman correlation, a non-parametric rank-based metric, is particularly effective for identifying non-linear dependencies in sparse and noisy data, which are common in ambient sensor networks. We demonstrate the accuracy of our approach on the Aruba dataset from the CASAS project, a real-world smart home dataset. The suggested approach outperforms widely used baselines including Pearson correlation, sensor-event windowing, and decision trees by 11.5% to 17.0%, with an average F1-score of 0.854. These findings confirm that correlation-driven dynamic segmentation is a promising strategy for achieving accurate and efficient HAR in real-time smart home applications.