Scholarlink

<p>The proliferation of fake profiles on social media platforms represents a critical security challenge, facilitating the spread of misinformation and malicious activities that undermine user trust. While numerous machine learning (ML) and deep learning (DL) models have been proposed for detection, the existing literature often lacks rigorous, comparative benchmarks on standardized, feature-based datasets, making it difficult to assess the relative efficacy of different approaches. This study addresses this gap by conducting a systematic performance analysis of a wide spectrum of models on a public Instagram dataset. The methodology includes a robust preprocessing pipeline and the evaluation of traditional classifiers (e.g., KNN, Random Forest), their hypertuned counterparts, and advanced deep learning architectures (DNN, LSTM, and a custom ResNet), all evaluated using metrics including accuracy and F1-score. Our experiments reveal that while optimized traditional models perform well, a ResNet-based architecture is exceptionally effective, achieving a state-of-the-art accuracy of 94.1%. This work establishes a valuable and reproducible performance benchmark and highlights the significant potential of applying architectures with residual connections to learn complex patterns in structured social media data, providing a clear direction for developing more robust detection 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

Deep vision against deception using CNN strategies for fake social media profile detection

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月11日

查看全文
  • Ajay Kumar,
  • Pranav B. Samant,
  • Swaraj Singh Negi,
  • Dhruv Pradeep Chhajed,
  • Praneet Saurabh

摘要

The proliferation of fake profiles on social media platforms represents a critical security challenge, facilitating the spread of misinformation and malicious activities that undermine user trust. While numerous machine learning (ML) and deep learning (DL) models have been proposed for detection, the existing literature often lacks rigorous, comparative benchmarks on standardized, feature-based datasets, making it difficult to assess the relative efficacy of different approaches. This study addresses this gap by conducting a systematic performance analysis of a wide spectrum of models on a public Instagram dataset. The methodology includes a robust preprocessing pipeline and the evaluation of traditional classifiers (e.g., KNN, Random Forest), their hypertuned counterparts, and advanced deep learning architectures (DNN, LSTM, and a custom ResNet), all evaluated using metrics including accuracy and F1-score. Our experiments reveal that while optimized traditional models perform well, a ResNet-based architecture is exceptionally effective, achieving a state-of-the-art accuracy of 94.1%. This work establishes a valuable and reproducible performance benchmark and highlights the significant potential of applying architectures with residual connections to learn complex patterns in structured social media data, providing a clear direction for developing more robust detection systems.