Presented By O’Reilly and Cloudera
Make Data Work
March 5–6, 2018: Training
March 6–8, 2018: Tutorials & Conference
San Jose, CA

Deep credit risk ranking with LSTM

Kyle Grove (Teradata)
1:50pm2:30pm Wednesday, March 7, 2018
Secondary topics:  Graphs and Time-series
Average rating: *****
(5.00, 5 ratings)

Who is this presentation for?

  • Data scientists

What you'll learn

  • Learn how companies in the financial industry have leveraged TensorFlow and Spark to productionize credit risk analytics


Kyle Grove explains how Teradata and some of world’s largest financial institutions are innovating credit risk ranking with deep learning techniques and AnalyticOps. With the AnalyticOps framework, these organizations have built models with increased accuracy to drive more profitable lending decisions while being explainable to regulators.

Topics include:

  • Overview of a machine learning ensemble including LSTM that achieves 90%+ accuracy at predicting delinquency/default, exceeding conventional credit risk methods by more than 20%
  • Overview of a model management accelerator that is used to build and deploy the models in an integrated cloud platform, based on TensorFlow and Spark, and supports Keras, Deeplearning4j and SparkML models
  • Overview of an innovative technique for model interpretability that obviates LIME’s need to generate synthetic examples
Photo of Kyle Grove

Kyle Grove


Kyle Grove is the chief data scientist for Teradata’s Wells Fargo relationship, where he employs his dual background in natural language processing and cognitive science to architect data science solutions that optimize banking functions in risk, compliance, service, and marketing. The productionalized solutions utilize machine learning at scale to predict and nudge human behavior to ends favorable to the bank and its customers.

Comments on this page are now closed.


Jonathan Bouchet | DATA SCIENTIST
03/09/2018 1:44am PST

Hi Kyle,
thanks for the very insightful talk. Will the slides be available online ? There were some references to model/technique that I didn’t had time to note during the talk.