How can I apply TensorFlow deep learning algorithms to data within Splunk for credit card fraud?

[akinobafemi]

Does anyone have an idea as to how I can apply deep learning algorithms (TensorFlow) to data within Splunk? I'm trying to do this for credit card fraud database at my bank and the data I would be working on is similar to the one obtainable via this link:
https://www.kaggle.com/dalpozz/creditcardfraud

Suggestions would be highly appreciated.

[gesman_splunk · Accepted solution]

Hello,

1.
Currently Splunk does not have direct integration with Tensorflow.
However you can easily get access to Splunk datasets within your Deep Learning solution either by leveraging Splunk API:
http://docs.splunk.com/Documentation/SplunkCloud/latest/RESTTUT/RESTsearches

or by leveraging Splunk SDK's that are available for multiple languages (with Python being first class citizen of course):
http://dev.splunk.com/sdks

Decoupling Splunk deployment from Deep Learning frameworks deployment will make sense in production for scalability as well as due to the need to access GPU hardware directly by DL side of things to train the models.

2.
We've done similar anti-fraud research project recently based on similar architectures: Splunk SDK/API + TensorFlow + Keras:
https://www.splunk.com/blog/2017/04/18/deep-learning-with-splunk-and-tensorflow-for-security-catchin...

3.
I'd be very interested to help you in this effort wherever possible. Please contact me directly: gesman at splunk.com

[koshyk]

hi, please can you put the details into a github project so we can all contribute too. cheers

[akinobafemi]

@gesman...thanks for your answer. I've sent you an email and hopefully we can work on an even improved solution and post here for the benefit of others.