Mike McPhee Anderson FSA, MS
Mike McPhee Anderson is an accomplished data scientist specializing in insurance and reinsurance. With a strong foundation in quantitative disciplines, he holds dual bachelor's degrees in Computer Science and Statistics, and a master's degree in Computer Science with an emphasis on Artificial Intelligence.
As a Fellow of the Society of Actuaries (FSA), Mike brings a unique blend of technical expertise and actuarial insight to his work. His career spans diverse roles, including software engineering, actuarial pricing, and actuarial valuation, equipping him with broad communication abilities and a strong understanding of various stakeholder perspectives.
Mike's professional focus lies in leveraging advanced analytics, data engineering, and machine learning to solve complex challenges and drive innovation in the reinsurance space. He is a life-long learner, that values soft-skills as much as technical skills.
As a former full-stack software engineer, Mike is exceptionally skilled with Linux, Docker, databases, and many programming languages and frameworks/technologies. In his current role, he works with Python, R, MLFlow, git, Spark, Snowflake, DBT, and Tableau.
Projects
Perm Mortality
The Colab project is a small slice of a much larger GitHub project. Poisson regression is the cornerstone of a lot of my professional work, since it allows estimation of claim rates on insurance policies.
The underlying data (ILEC 2009-16) is from the Society of Actuaries (SOA). The data presents some interesting modeling challenges due to the variety of different policy types that are intermingled and need to be teased apart using some domain knowledge.
Colab Notebook (Python / R)
GitHub Project (R)
San Francisco Fire Station Modeling
Some of the most interesting (and most plentiful) public data is geospatial in nature. San Francisco publishes a substantial amount of data, including granular data about police and fire calls.
This project looks at fire department response times and tries to identify a location for a (hypothetical) new fire station that would lower response times in areas where responses are typically over 5 minutes. Response times are estimated / smoothed using a kernel density estimator (KDE), and then converted to polygons for use by the web-based tool. The impacts to response times for the simulated new fire station location is based on data pulled from Bing Maps.
Web-based Viz (d3.js)
Experience
Reinsurance Group of America
- Led projects from inception to implementation, including commercially available predictive models (MedScore). Worked with many stakeholders in actuarial, underwriting, leadership, and I.T. to develop both a product vision and implementation strategy.
- Developed internal risk-scoring predictive models using Spark, Python, R, DBT, and SQL. Methods include gradient boosted trees (XGboost), decision trees, and generalized linear models (GLMs).
- In close collaboration with assumptions actuaries, developed predictive models and validation tools for lapse and mortality assumptions for the U.S. Individual Life reinsurance block. Earned the trust needed to apply predictive models to mortality business with $10bn+ in annual claims.
- Implemented best practices around model version control using MLFlow and github. Mentored junior staff on predictive modeling, data and feature engineering, business topics, and soft skills.
Securian
Allianz Life
Garmin
C.H. Robinson
Education
Georgia Tech
GPA: 4.0