MortgageMachineLearning

This repository is the rough working code for the GOTO CPH 2015 talk by Phil Winder title Modern Fraud Prevension using Deep Learning. It was produced in a rush, so probably has some issues.

Requirements

DB requirements

postgres

Python requirements

Entire Scipy stack sklearn keras pandas pg8000

Examples

MNIST digits example use

Run KerasMNISTTest.py. It should download the data and produce outputs in a ./plots/ folder.

Speaker example use

Speaker data is available from: http://web.mit.edu/6.863/share/nltk_lite/timit/

To run the code, start by running speakerPreprocess.py, to compute the STFT of all the audio data. Then, to perform the classification, run speakerDeepLearning.py.

Mortgate example use

Mortgate data is available from: http://www.freddiemac.com/news/finance/sf_loanlevel_dataset.html

See the loader files for some tips.

Start by loading the database with the data using the load_raw_mac or load_raw_ubuntu files. Then run the classification code in MortgageDeepLearning.py. It should generate some plots in the ./plots/ folder

Mortgage Decision Tree example

Add the data as per Mortgage Deep Learning instructions. Run MortgageRandomForest.py. Outputs should be in the ./plots/ folder.

Acknowledgements

This code was based upon code from Todd Schneider, many thanks. https://github.com/toddwschneider/agency-loan-level

Thanks also to Python, SKLearn, Keras, PG8000 devs. Ta very much.