Scholarlink

<p>Amidst the era of infrastructure-less sensing, the domain of human activity recognition (HAR) experiences a substantial evolution by harnessing the ubiquity of sensors in smartphones. However, the constrained computational capabilities of handheld devices pose challenges in meeting the real-time responsiveness demands of HAR applications, which necessitates the emergence of Feature Selection (FS) techniques for dimensionality reduction for on-device learning and real-time activity prediction. Considering the data-intensive nature of HAR applications, this paper explores the application of Genetic Algorithms (GAs) for feature selection on smartphone-based sensor data, focusing on identifying the most informative and relevant features to improve the efficiency of HAR models. Experimental results are shown for the benchmark sensor dataset. The identification of common features, distinct classifier behavior, and feature minimization underscore the importance of feature selection in enhancing model efficiency and interpretability. The proposed training pipeline with cross-validation is found to achieve around 95% accuracy with only 20 important features as compared to around 78% accuracy for all features.</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 feature selection approach for smartphone-based human activity recognition applying genetic algorithm

  • 期刊论文

  • 发表日期：

    2025年4月1日

查看全文
  • Anindita Saha,
  • Srijeeta Guha,
  • Chandreyee Chowdhury

摘要

Amidst the era of infrastructure-less sensing, the domain of human activity recognition (HAR) experiences a substantial evolution by harnessing the ubiquity of sensors in smartphones. However, the constrained computational capabilities of handheld devices pose challenges in meeting the real-time responsiveness demands of HAR applications, which necessitates the emergence of Feature Selection (FS) techniques for dimensionality reduction for on-device learning and real-time activity prediction. Considering the data-intensive nature of HAR applications, this paper explores the application of Genetic Algorithms (GAs) for feature selection on smartphone-based sensor data, focusing on identifying the most informative and relevant features to improve the efficiency of HAR models. Experimental results are shown for the benchmark sensor dataset. The identification of common features, distinct classifier behavior, and feature minimization underscore the importance of feature selection in enhancing model efficiency and interpretability. The proposed training pipeline with cross-validation is found to achieve around 95% accuracy with only 20 important features as compared to around 78% accuracy for all features.