Scholarlink

<p>This study proposes an automatic learning activity classification framework based on the Tri Pramana concept using immersive Virtual Reality (VR) video data. Tri Pramana comprising, Sabda, Pratyaksa, and Anumana, is operationalized in this research as a set of observable learning activity categories rather than as a comprehensive epistemological model. Video data were collected from a Fiber Optic Splicing practicum conducted in an immersive VR environment, with each video clip segmented into 5-second sequences. A total of 1,036 video clips were used, consisting of 824 training samples (80%) and 212 testing samples (20%). To evaluate classification performance, a comparison was conducted between Long-term Recurrent Convolutional Network (LRCN) and Convolutional Long Short-Term Memory (ConvLSTM) architectures. Experimental results indicate that the LRCN model outperforms ConvLSTM, achieving an accuracy of 94.34, precision of 94.12, recall of 95.00, F1-score of 94.56, and an inference time of 0.897&#xa0;s. The optimal hyperparameters include a learning rate of 1e − 4, batch size of 32, and maximum sequence length of 20. Beyond technical performance, the proposed classification framework provides a foundation for learning analytics in immersive VR environments. By translating observable learning activities into interpretable behavioral indicators, the model has potential applications in monitoring learner engagement and supporting reflective feedback in practicum-based learning contexts.</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

LRCN-based Tri Pramana learning activity classification from immersive virtual reality videos

  • 期刊论文

  • 发表日期：

    2026年5月1日

查看全文
  • I Gede Partha Sindu,
  • Made Sudarma,
  • Agus Aan Jiwa Permana,
  • Putu Zasya Eka Satya Nugraha

摘要

This study proposes an automatic learning activity classification framework based on the Tri Pramana concept using immersive Virtual Reality (VR) video data. Tri Pramana comprising, Sabda, Pratyaksa, and Anumana, is operationalized in this research as a set of observable learning activity categories rather than as a comprehensive epistemological model. Video data were collected from a Fiber Optic Splicing practicum conducted in an immersive VR environment, with each video clip segmented into 5-second sequences. A total of 1,036 video clips were used, consisting of 824 training samples (80%) and 212 testing samples (20%). To evaluate classification performance, a comparison was conducted between Long-term Recurrent Convolutional Network (LRCN) and Convolutional Long Short-Term Memory (ConvLSTM) architectures. Experimental results indicate that the LRCN model outperforms ConvLSTM, achieving an accuracy of 94.34, precision of 94.12, recall of 95.00, F1-score of 94.56, and an inference time of 0.897 s. The optimal hyperparameters include a learning rate of 1e − 4, batch size of 32, and maximum sequence length of 20. Beyond technical performance, the proposed classification framework provides a foundation for learning analytics in immersive VR environments. By translating observable learning activities into interpretable behavioral indicators, the model has potential applications in monitoring learner engagement and supporting reflective feedback in practicum-based learning contexts.