AI-Driven Detection of Anomalous Click Patterns in Online Advertising

• Nikhil Reddy Pallepati

  Author

Click fraud is still a major issue in online advertising that results in thousands of dollars lost by advertisers, diluting the effectiveness of advertising platforms. A group of researchers in 2023 applied machine learning and ensemble learning to analyze a framework that identifies outlier clicks in mobile advertising. The FDMA 2012 benchmark dataset, released by BuzzCity, contains large-scale real-world clickstream records characterized by severe class imbalance and noisy behavioral patterns and is used to evaluate the proposed methodology. To enhance fraud detection, the framework will extract temporal, behavioral, and device-related features, including click velocity, burstiness, off-peak activity, user-agent entropy, inter-click interval statistics, and more. The counterfeit and original samples’ imbalance is treated by the Synthetic Minority Over-sampling Technique (SMOTE). After that, a stacking ensemble architecture, which combines the random forest and XGBoost and then a multi-layer perception is trained for classification. As shown by the experimental results, the ensemble produces better classification results when compared to any single classifier and rule-based methods, with an F1-score of 0.91 and AUC-ROC of 0.96. According to the results, it is possible to achieve accurate click fraud detection that scales to near real-time for online advertising systems with fine-grained temporal feature engineering and ensemble learning.

• PDF

Copyright (c) 2026 International Journal of Intelligent Systems and Data Science

This work is licensed under a Creative Commons Attribution 4.0 International License.