ellisvalentiner / credit-card-fraud Goto Github PK

Hi Ellis,

First, thanks for putting your work on GitHub - it has been a challenge to train a neural network on the credit card fraud data set, since most of the networks I've tried insist on being "lazy" and classifying everything as not fraud. It's great to have a working example as a starting point to figure out how to improve what I'm doing.

I was wondering how I can describe the implementation of your network elsewhere (e.g., my data-engineering-scenarios/kaggle-sql-jupyter-keras repository, which is illustrating an SQL-to-Keras data pipeline). Did the SReLU architecture/layer idea originally come from you, or was it mentioned in the Dal Pazollo paper or some other paper?

Of course, your work in implementing it in Keras was ultimately the crucial ingredient, and the important step for helping me get some traction on the problem - so thanks again!

It would be useful to visualize the network weights

i would like to understand what the numbers in the csv file mean as i can only see time amount and class.

i would like to add or modify the transaction results to integrate with the data i have collected and add geo ip and device fingerprint to the equation

thank you so much