C2IDTL: Novel Click to Image Conversion Approach for Deep Transfer Learning in Click Fraud Detection on Digital Platforms
The global retail industry has witnessed a massive expansion of e-commerce businesses in the last decade by transforming physical stores into digitalised entities. One of the biggest threats to e-commerce is click fraud, which generates fraudulent clicks without genuine intention on online advertise...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2025-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/11006638/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The global retail industry has witnessed a massive expansion of e-commerce businesses in the last decade by transforming physical stores into digitalised entities. One of the biggest threats to e-commerce is click fraud, which generates fraudulent clicks without genuine intention on online advertisements using automated software programs or organised human activities. This has become a major threat to the online advertising eco systems, causing heavy losses for tech companies like Google, Facebook, and Meta which rely on pay per click (PPC) models. To overcome click fraud, we propose a novel click to image conversion technique to generate synthetic images from clicks and perform click fraud detection on those generated images directly with Xception, Inception, VGG, ResNet and DenseNet DTL models (C2IDTL). This novel approach embedded the feature importance and spatial structure into synthetic images to enable the identification of complex pattern recognition via DTL models. We evaluated the performance of C2IDTL on three publicly available datasets, namely FDMA 2012 BuzzCity, TalkingData AdTracking and Sweden Ad click datasets and compared the performances with the conventional state of the art ML models. C2IDTL enhances the fraud detection by exceeding 91.18% accuracy, 90.90% F1 score and 96.45% AUC based on the k-fold cross validation performed on all three datasets. Consequently, these results validate the robustness and the ability to generalise across diverse click datasets evaluated in prior scientific literature. Despite DTL’s association with the visual domain, this study challenges its traditional usage and opens avenues for future studies to optimise performance and robustness. |
|---|---|
| ISSN: | 2169-3536 |