Ad Click Fraud Detection Using Machine Learning and Deep Learning Algorithms

Alzahrani, Reem A.

doi:10.1109/ACCESS.2025.3532200

2025-01-01

Alzahrani, Reem A. *

Malak Aljabri

Rami Mustafa A. Mohammad

DOI：10.1109/ACCESS.2025.3532200

原文链接

原文求助

摘要

En 中文

In online advertising, click fraud poses a significant challenge, draining budgets and threatening the industry's integrity by redirecting funds away from legitimate advertisers. Despite ongoing efforts to combat these fraudulent practices, recent data emphasizes their widespread and persistent nature. Toward detecting click fraud effectively, this study employed a comprehensive feature engineering and extraction approach to identify subtle differences in click behavior that could be used to distinguish fraudulent from legitimate clicks. Subsequently, a thorough evaluation was conducted involving nine diverse machine learning (ML) and Deep Learning (DL) models. After Recursive Feature Elimination (RFE), the ML models consistently demonstrated robust performance. DT and RF surpassed 98.99% accuracy, while GB, LightGBM, and XGBoost achieved 98.90% or higher. Precision scores, measuring accurate identification of fraudulent clicks, exceeded 98% for models like ANN. In parallel, deep learning (DL) models, including Convolutional Neural Network (CNN), Deep Neural Network (DNN), and Recurrent Neural Network (RNN), showcased strong performance. RNN, in particular, achieved 97.34% accuracy, emphasizing its efficacy. The study underscores the prowess of tree-based methods and advanced algorithms in detecting click fraud, as evidenced by high accuracy, precision, and recall scores. These findings contribute valuable insights to combat click fraud and establish the groundwork for the strategic development of anti-fraud measures in online advertising.

Keyword:

Fraud

Feature extraction

Accuracy

Chatbots

Advertising

Deep learning

Classification tree analysis

Boosting

Random forests

Nearest neighbor methods

Click fraud