Scholarlink

This chapter addresses the challenges of HAR in uncontrolled environments by proposing a variable batch size-based deep learning framework. Traditional HAR systems often struggle in real-world scenarios due to data mismatches and computational limitations. We introduce a novel approach using a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model, coupled with a variable batch size strategy that dynamically adjusts during training to enhance model generalization. A custom dataset, collected from smartphone sensors in uncontrolled settings, is utilized to better represent real-world activities. Our framework demonstrates superior performance, achieving an average performance accuracy of 98% compared to benchmark models while effectively managing computational resources. This research highlights the efficacy of variable learning strategies in improving HAR robustness and applicability in dynamic environments.

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 Variable Batch Size-Based Hybrid Deep Learning Framework for HAR in Uncontrolled Environments

  • 章节

  • 发表日期：

    2026年

查看全文
  • Nurul Amin Choudhury,
  • Badal Soni

摘要

This chapter addresses the challenges of HAR in uncontrolled environments by proposing a variable batch size-based deep learning framework. Traditional HAR systems often struggle in real-world scenarios due to data mismatches and computational limitations. We introduce a novel approach using a hybrid Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) model, coupled with a variable batch size strategy that dynamically adjusts during training to enhance model generalization. A custom dataset, collected from smartphone sensors in uncontrolled settings, is utilized to better represent real-world activities. Our framework demonstrates superior performance, achieving an average performance accuracy of 98% compared to benchmark models while effectively managing computational resources. This research highlights the efficacy of variable learning strategies in improving HAR robustness and applicability in dynamic environments.