Click Fraud Detection Using Machine Learning

Abstract

It could be challenging to identify dishonest publishers in the online pay-per-click advertising space because there are fewer fraudulent cases than successful ones. This mismatch leads to incorrect data classification by automated learning systems that are biased towards the dominant class. When user click datasets are not made available to the public, click fraud detection becomes more challenging. The phenomenon when publications show consistent behavioral trends over time but their actual labels change, exacerbates the issue. The solution to this issue lies in the creation of fraud detection systems that can adapt to changing behavior without requiring complete retraining. The TalkingData Ad Tracking Fraud Detection (TDA) dataset is used in this study to get around these problems. The results of the deployment show that it can recognize clicks.