OCAN: One-Class Adversarial Nets for Fraud Detection

In this paper, we develop one-class adversarial nets (OCAN) for fraud detection with only benign users as training data.

Running Environment

The main packages you need to install are listed as follow

1. python 2.7 
2. tensorflow 1.3.0

DateSet

For experiments, we evaluate OCAN on two real-world datasets: wiki and credit-card which have been attached in folder data/.

Model Evaluation

The command line for OCAN goes as follow

    python oc_gan.py $1 $2

where $1 refers to different datasets with wiki 1, credit-card(encoding) 2 and credit-card(raw) 3; $2 denotes whether some metrics, such as fm_loss and f1 in training process, are provided, with non-display 0 and display 1.

   e.g. python oc_gan.py 1 0

The above command line shows the performance of OCAN on wiki without displaying metrics in the training process.

Authors

Panpan Zheng
- personal website
- google scholar

Citation

I am very glad that you could visit this github and check my research work. If it benefits your work, please refer this work by

@article{zheng2018one,
  title={One-Class Adversarial Nets for Fraud Detection},
  author={Zheng, Panpan and Yuan, Shuhan and Wu, Xintao and Li, Jun and Lu, Aidong},
  journal={arXiv preprint arXiv:1803.01798},
  year={2018}
}

Acknowledgments

This work was going on underlying the guide of prof. Xintao Wu and Dr. Shuhan Yuan.

Appreciate it greatly for every labmate in SAIL lab

georgy95 / OCAN