SAFE: A Neural Survival Analysis Model for Fraud Early Detection

In this paper, we propose a survival analysis based fraud early detection model, SAFE, that maps dynamic user activities to survival probabilities that are guaranteed to be monotonically decreasing along time.

Running Environment

The main packages you need to install

1. python 2.7 
2. tensorflow 1.3.0

DateSet

For experiments, we evaluate SAFE on two real-world datasets: twitter and wiki which have been attached in twitter/ and wiki/, respectively.

Model Evaluation

The command lines for SAFE and baselines go as follow

• SAFE

    python framework/safe.py $1

• M-LSTM

    python framework/base_rnn.py $1

• CPH & SVM

    python framework/safe_baselines.py $1

where $1 refers to datasets on which the model runs, and it can be assigned as 'twitter' or 'wiki'.

• Weibull & other distributions

    python framework/safe_distr.py $1 $2

where $1 refers to the corresponding distributions and it can be assigned as 'exp' (exponential), 'ray' (Rayleigh) and 'poi' (poisson); $2 denotes the datasets, 'twitter' or 'wiki'.

Authors

• Panpan Zheng, Shuhan Yuan and Xintao Wu

  • 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 cite the paper in Arxiv https://arxiv.org/abs/1809.04683v1 .

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