Scholarlink

To allow instructors to reach out and assist students in a timely manner, a student success prediction model aims to identify which students could use help. We focus on hands-on cybersecurity education and the EDURange platform. Because it is difficult to obtain sufficient high-quality data to train the model exclusively through classroom testing, we are employing the 3DG framework to create synthetic data to supplement it. This paper is an in-depth exploration of the tensor decomposition aspect of the 3DG methodology for our use-case by offering a thorough explanation of tensor decomposition, creating a new optimization strategy, and exploring the process on our new dataset. We found that the existing 3DG optimization produced a zero-tensor result, which cannot be used to generate new data points. We make use of the likelihood that if a student answers a question correctly on an early attempt, they are likely to answer it correctly on all subsequent attempts.

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

Tensor Decomposition Optimization for Student Success Prediction Modeling in Hands-on Cybersecurity Exercises

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Julia Scott,
  • Jens Mache,
  • Richard Weiss,
  • Taylor Wolff,
  • Jack Cook

摘要

To allow instructors to reach out and assist students in a timely manner, a student success prediction model aims to identify which students could use help. We focus on hands-on cybersecurity education and the EDURange platform. Because it is difficult to obtain sufficient high-quality data to train the model exclusively through classroom testing, we are employing the 3DG framework to create synthetic data to supplement it. This paper is an in-depth exploration of the tensor decomposition aspect of the 3DG methodology for our use-case by offering a thorough explanation of tensor decomposition, creating a new optimization strategy, and exploring the process on our new dataset. We found that the existing 3DG optimization produced a zero-tensor result, which cannot be used to generate new data points. We make use of the likelihood that if a student answers a question correctly on an early attempt, they are likely to answer it correctly on all subsequent attempts.