Scholarlink

Fraudulent activities, particularly click fraud, present a significant challenge in online advertising, negatively impacting advertisers as well as adverting platforms. As attackers continuously evolve their strategies to bypass existing detection mechanisms, there is a pressing need for adaptive techniques capable of identifying patterns and determining appropriate actions. This research examines the use of artificial intelligence (AI) classifier-based detection, particularly leveraging long short term memory (LSTM) models optimized by modified metaheuristic algorithms. A modified variant of the recently developed elk herd optimization (EHO) algorithm, which incorporates an adaptive hill climbing strategy, is introduced to address the constraints of the baseline method and applied for hyperparameter tuning of LSTM training and architecture parameters. The proposed approach is evaluated and a comparative analysis carried out in a set of simulations to determine suitable LSTM parameters that demonstrated favorable outcomes for click fraud detection. Simulations on a real-world dataset demonstrate promising results, with the highest-performing model obtaining an accuracy of 0.792701.

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

Click Pattern Classification Using Modified Metaheuristic Optimized Long-Short Term Neural Networks for Fraud Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Lepa Babic,
  • Stefan Ivanovic,
  • Luka Jovanovic,
  • Marija Markovic Blagojevic,
  • Vico Zeljkovic,
  • Miodrag Zivkovic,
  • Branislav Radomirovic,
  • Nebojsa Bacanin

摘要

Fraudulent activities, particularly click fraud, present a significant challenge in online advertising, negatively impacting advertisers as well as adverting platforms. As attackers continuously evolve their strategies to bypass existing detection mechanisms, there is a pressing need for adaptive techniques capable of identifying patterns and determining appropriate actions. This research examines the use of artificial intelligence (AI) classifier-based detection, particularly leveraging long short term memory (LSTM) models optimized by modified metaheuristic algorithms. A modified variant of the recently developed elk herd optimization (EHO) algorithm, which incorporates an adaptive hill climbing strategy, is introduced to address the constraints of the baseline method and applied for hyperparameter tuning of LSTM training and architecture parameters. The proposed approach is evaluated and a comparative analysis carried out in a set of simulations to determine suitable LSTM parameters that demonstrated favorable outcomes for click fraud detection. Simulations on a real-world dataset demonstrate promising results, with the highest-performing model obtaining an accuracy of 0.792701.