Scholarlink

<p>The rapid growth of online learning has intensified the need for reliable remote proctoring systems to ensure academic integrity. Detecting suspicious behaviours such as impersonation and cheating remains a challenging task due to the subtle nature of many illicit actions. In this paper, an attention-enhanced deep learning framework is proposed for automated suspicious behaviour detection in online examinations. The method is based on transfer learning using pretrained VGG16 and VGG19 networks integrated with the Convolutional Block Attention Module (CBAM) to emphasize discriminative spatial and channel-wise features related to abnormal behaviour. A custom dataset was developed from controlled real-world examination scenarios to train and evaluate the models. To improve convergence and detection performance, RMSprop and Adam optimizers were employed. Experimental results demonstrate that the proposed system achieves a detection accuracy of 88.33% with CBAM-VGG16 (RMSprop) and 90.00% with CBAM-VGG19 (Adam)<b>.</b> These results confirm the effectiveness of attention-guided deep learning in enhancing the security and reliability of remote proctoring systems.</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 convolutional block attention module-enhanced VGG architecture for robust suspicious activity detection in online examinations

  • 期刊论文

  • 发表日期：

    2026年7月23日

查看全文
  • Arunendra Singh,
  • Himanshu Kumar Diwedi,
  • Mohit Kumar Srivastava,
  • Ashish Kumar Chakraverti

摘要

The rapid growth of online learning has intensified the need for reliable remote proctoring systems to ensure academic integrity. Detecting suspicious behaviours such as impersonation and cheating remains a challenging task due to the subtle nature of many illicit actions. In this paper, an attention-enhanced deep learning framework is proposed for automated suspicious behaviour detection in online examinations. The method is based on transfer learning using pretrained VGG16 and VGG19 networks integrated with the Convolutional Block Attention Module (CBAM) to emphasize discriminative spatial and channel-wise features related to abnormal behaviour. A custom dataset was developed from controlled real-world examination scenarios to train and evaluate the models. To improve convergence and detection performance, RMSprop and Adam optimizers were employed. Experimental results demonstrate that the proposed system achieves a detection accuracy of 88.33% with CBAM-VGG16 (RMSprop) and 90.00% with CBAM-VGG19 (Adam). These results confirm the effectiveness of attention-guided deep learning in enhancing the security and reliability of remote proctoring systems.