Balancing ML accuracy, interpretability and costs when building a model

Abstract: As data scientists we strive to deliver high performance models, but in the real-world the best model possible is not usually the best model for the business. When developing a model if it is not interpretable by the business, you will be unable to get buy in necessary to get your model into production. Additionally, you are always fighting two cost related battles: opportunity cost of delivering a perfect model tomorrow instead of delivering a good one today; operational costs of the most superior model compared to the next best one. This workshop will use real-world coding examples in Python to demonstrate how to be mindful of these constraints when developing your models.

Bio: Marc Fridson is the Principal Data Scientist of Cross Brand Digital @ Carnival Cruise Line, a Part-Time Lecturer for the Applied Analytics Program Masters Program @ Columbia University and the founder of tech start-up Instant Analytics.

Marc has previously worked as a Technology Consultant for Accenture, as an Engineer for the Boeing Company, AVP of Metrics and Reporting for Capital One, and as Manager of Analytics at CB Richard Ellis for JP Morgan Chase’s Real Estate Management. Previous consulting clients include: Morgan Stanley, Capital One, The College Board, Anthem Blue Cross, Verizon and Time Warner Cable.

He has helped these companies measure, analyze, and automate their processes through data analysis and by developing technological tools to enable process improvement/automation.

He holds a B.S. in Industrial and Systems Engineering from Rutgers University.