Courses: How to Increase Lending to Women Using Machine Learning and Reject Inference
Course Description
This open-source 6-hour video course equips both product owners and data scientists with the tools they need to address bias within credit evaluation processes at financial service providers. Within the course, participants learn to deploy algorithmic solutions to rectify biases in credit evaluations, specifically focusing on the high-impact issue of mistakenly rejecting creditworthy applicants, often due to biases ingrained in evaluation models themselves. Users will walk away with:
- Working knowledge of where gender balance vulnerabilities emerge across credit processes
- The language needed to discuss fair lending with colleagues, in any legal or policy environment
- Two innovative methods to directly address bias in credit evaluation processes:
- Matching algorithms
- Counterfactual learning
- Working knowledge of Women’s World Banking’s tested approach to assessing bias and balance at various stages in a credit evaluation process, with a particular focus on data and decision-making engines
- Real-world examples and insights from credit evaluation balance audits with financial institutions spanning three continents
Course Outline
The first section (Section I) of this course is designed for product owners and data scientists alike, with no prior knowledge or technical background required. This section conveys the foundational knowledge necessary to understand and address bias against rejected applicants. It covers core concepts such as credit evaluation, bias in credit evaluation, the meaning of fairness in this context, algorithmic bias, and reject inference.
The second section (Section II) requires technical expertise and is suited for those familiar with machine learning concepts. We will introduce matching algorithms, explaining how to integrate these with machine learning to identify creditworthy applicants who were erroneously rejected. Additionally, the course will cover counterfactual learning and its applications in reject inference.
By the end of this course, participants will have working knowledge of credit evaluation, bias in credit evaluation, fair lending, and the application of machine learning to foster a fair credit evaluation process.
Course Content
Section I
Course Introduction
Foundational Concepts
Guest Speaker: Message from Dr. Sonja Kelly
Detecting and Mitigating Gender Biases within Credit Evaluation Processes
Introduction to Reject Inference
Guest Speaker: Message from Pallavi Madhok
Section II
Module I
Reject Inference Via Matching Algorithms
Constructing Matching Scores I
Constructing Matching Scores II
Reject Inference Via Matching Algorithms in Python
Module II
Counterfactual Learning and its Applications in Reject Inference
Reject Inference Through Noisy Label Detection and Counterfactual Correction
Acknowledgments
About the Instructor
Mehi (Mehrdad) Mirpourian, Data Science Manager at Women’s World Banking, possesses extensive expertise in machine learning and statistical learning, focusing on creating inclusive financial solutions through innovative product design.
Mehi has partnered with financial institutions to transform their challenges into algorithmically solvable problems, while also collaborating with academic and research institutes for methodological rigor.
His primary goal is using data science to foster fairer customer conditions and reduce missed opportunities for financial service providers, underpinning the belief that fairness is key to mutual prosperity and economic empowerment.
Mehi holds an MS in Statistics from Baruch College, an MS in Management Engineering from the Polytechnic University of Milan, and a BS in Industrial Engineering and Optimization from K.N. Toosi University of Technology.
Acknowledgments
This course would not have been possible without the efforts of many individuals who contributed to the ideas, the content, the production, and the resources necessary to ensure is success.
Course content was developed in partnership with State University of New York at Albany. Dr. Charalampos Chelmis and PhD candidates Sadia Rahman and Mahsa Azarshab contributed expertise that enhanced the rigor of the methods we jointly developed for reject inference.
Within Women’s World Banking, Dr. Megan Dwyer Baumann aided in curriculum development; Chelsea Bennington and Arianne Wunder worked on the communications aspects of the course; Molly Burns ensured the project continued to be aligned with its goals; Dr. Sonja Kelly and Pallavi Madhok provided feedback, and video producer Dakota Lupo led on production.
This publication is made possible by PayPal as part of its support of Women’s World Banking’s work using data science to increase financial services for low-income women globally.
