aflaisler / fraud_detection_testing

The fraud detection app:

The aim of the program is to flag potential new fraud for further review as it comes in so it can be triaged by most pressing (and costly) transactions.

How it works?

Given a JSON file with 137k transactions that are fraud or not, and other features, create a web based front-end with machine learning back end to enable quick triage of potential new fraud.

Modeling + Evaluation

After EDA, having high dimensional training set with categorical, numerical and discret values, the app is based on the RandomForest Classifier. Once trained (GridSearched + CrossValidated) the model is stored using cPickle (the c version of Pickle allows a loading time < 3sec.) and reloaded and each GET/POST requests.

Final performance of the model:

F1 score: 0.59, Precision: 0.93, Recall: 0.43, Accuracy: 0.95

Web app features:

Registers service at POST /register. Accepts input records on POST /score endpoint. Gives triage label: low, medium, high risk. Record predictions with attribute from the feature vectors in a postgresql db. Host the app on AWS EC2.