Scholarlink

Analysis of human behaviour in IoT applications based on human interaction is the area in which human activity recognition has drawn a lot of attention. In this paper, we proposed an intelligent model integrating multivariate dynamic mode decomposition (MDMD) and ensemble machine learning model to recognise physical human activity. The sensor data is decomposed into dynamic modes using MDMD. Different features including statistical features, power, average absolute amplitude, and frequency are derived from each mode to represent different classes of human activity. To classify the exacted features, several ensemble learning models are employed. The proposed model’s performance is evaluated using two datasets, UCI-HAR, and WISDM. The proposed model obtained remarkable accuracies of 97.6, and 95.5%, F1-score of 95%, and 93.20% for UCI-HAR, and WISDM respectively. Our findings proved that the proposed model is superior to competing previous models.

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

Robust Approach for Human Activity Recognition Using Decomposing Technique Based Machine Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Suha Zadain Hmoud,
  • Mohammed Diykh,
  • Shahab Abdulla,
  • Hussein Alabdally,
  • Aqeel Sahi

摘要

Analysis of human behaviour in IoT applications based on human interaction is the area in which human activity recognition has drawn a lot of attention. In this paper, we proposed an intelligent model integrating multivariate dynamic mode decomposition (MDMD) and ensemble machine learning model to recognise physical human activity. The sensor data is decomposed into dynamic modes using MDMD. Different features including statistical features, power, average absolute amplitude, and frequency are derived from each mode to represent different classes of human activity. To classify the exacted features, several ensemble learning models are employed. The proposed model’s performance is evaluated using two datasets, UCI-HAR, and WISDM. The proposed model obtained remarkable accuracies of 97.6, and 95.5%, F1-score of 95%, and 93.20% for UCI-HAR, and WISDM respectively. Our findings proved that the proposed model is superior to competing previous models.