Responsible Data Science Project: Evaluating Fairness and Transparency for Automated Decision System (Nutritional Labeling)
This project evaluates a top leaderboard automated decision system (ADS) solution from the Kaggle competition hosted by Home Credit for predicting default risk in potential new customers. The solutions for the Kaggle competition were scored using the AUC evaluation metric (the area under the curve of the true positive rate versus the false positive rate). In this project, we take apart our selected ADS, which is a solution based around using LightGBM, and additionally evaluate the ADS based on fairness in predictions, as well as the transparency and explainability of the model.
See: Project Report
See: Project Presentation