Skip to content

Submission for the FICO Explainable Machine Learning Challenge

Notifications You must be signed in to change notification settings

5teffen/FICO-xML-Challenge

Repository files navigation

Fico-Challenge

Explainable Machine Learning Challenge

This is the second place winning submission for the FICO Explainable Machine Learning Challenge created by Oscar Gomez and Steffen Holter. By combining instance level explanations and a general global model interpretation we have created an interactive application to visualize the logic behind each of the model’s decisions.

Individual Explanation

The local explanation visualizes each individual sample by showing the comparative values of each feature and their relative density distribution across the data set. Each explanation highlights the key features that the model used to make the decision. In addition, where applicable the solution suggests the necessary changes needed to reverse the decision.

Global Explanation

The global explanation acts as a way to aggregate the individual results across the whole data set. By either choosing to examine the key features or the necessary changes, the interactive hierarchy creates a comprehensive overview. Each click reveals an additional level of depth within the solution revealing more information.


Results Announcement

https://www.fico.com/en/newsroom/fico-announces-winners-of-inaugural-xml-challenge

Available Online at

http://www.ml-explainer.com/challenge_intro
http://nyuvis-web.poly.edu/projects/fico/intro

To run locally

  • Install python 3.6 or above
  • Pip install pandas, flask, numpy, scipy, sklearn
  • Open Webapp folder
  • Run hello.py
  • Open in localhost:5005

Suggested Specifications

  • Open on Mac or PC
  • Latest version of Google Chrome
  • Screen size of at least 13 inches

Contact us at [email protected] or [email protected] for any questions regarding the tool

About

Submission for the FICO Explainable Machine Learning Challenge

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published