Scholarlink

<p>With the rapid expansion of e-learning platforms, maintaining academic honesty during online assessments has become increasingly challenging. Traditional monitoring methods often fall short in detecting advanced cheating behaviors and may raise privacy concerns among students. To address these limitations, we propose LSTM-SWAP, a hybrid deep learning model that combines Long Short-Term Memory (LSTM) networks with a specialized attention mechanism called Sliding Window Attention Processor (SWAP) that focuses on key patterns in student behavior over time. This approach allows the model to highlight important behavioral cues while filtering out irrelevant data, thereby improving detection accuracy. Unlike conventional systems that rely on fixed features, LSTM-SWAP adapts dynamically to new and evolving cheating strategies, ensuring resilience against emerging threats. Additionally, the model is designed to reduce false alarms, promoting fairness by minimizing the risk of misidentifying honest students. We evaluated our model on a behavioral dataset based on mouse movement patterns using stratified fivefold cross-validation, achieving an average accuracy of 97.73%, F1-score of 97.74%, FPR of 0.0089, and an AUC-ROC of 0.9885, demonstrating its effectiveness for real-time cheating detection in online learning environments. These findings highlight the model’s potential as an flexible, scalable, and privacy-friendly solution for preserving academic integrity in digital learning platforms.</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

LSTM-SWAP: A Hybrid Deep Learning Model for Cheating Detection

  • 期刊论文

  • 发表日期：

    2025年9月4日

查看全文
  • Brahim Khalil Sedraoui,
  • Abdelmadjid Benmachiche,
  • Akram Bennour,
  • Amina Makhlouf,
  • Makhlouf Derdour,
  • Fahad Ghabban

摘要

With the rapid expansion of e-learning platforms, maintaining academic honesty during online assessments has become increasingly challenging. Traditional monitoring methods often fall short in detecting advanced cheating behaviors and may raise privacy concerns among students. To address these limitations, we propose LSTM-SWAP, a hybrid deep learning model that combines Long Short-Term Memory (LSTM) networks with a specialized attention mechanism called Sliding Window Attention Processor (SWAP) that focuses on key patterns in student behavior over time. This approach allows the model to highlight important behavioral cues while filtering out irrelevant data, thereby improving detection accuracy. Unlike conventional systems that rely on fixed features, LSTM-SWAP adapts dynamically to new and evolving cheating strategies, ensuring resilience against emerging threats. Additionally, the model is designed to reduce false alarms, promoting fairness by minimizing the risk of misidentifying honest students. We evaluated our model on a behavioral dataset based on mouse movement patterns using stratified fivefold cross-validation, achieving an average accuracy of 97.73%, F1-score of 97.74%, FPR of 0.0089, and an AUC-ROC of 0.9885, demonstrating its effectiveness for real-time cheating detection in online learning environments. These findings highlight the model’s potential as an flexible, scalable, and privacy-friendly solution for preserving academic integrity in digital learning platforms.