Naomi Adeyemi — Senior Data Scientist
Boston, MA
Data scientist with eleven years in healthcare and energy, working mostly on forecasts and risk models other teams depend on daily. Took a utility's day-ahead load forecast from 4.8% to 2.9% mean absolute percentage error, worth roughly $6.1m a year in avoided imbalance charges. Returned from a planned career break in 2023 and now sets the review standard for models that touch billing.
Experience
Senior Data Scientist, Granite Harbor Power · September 2023 – present
Forecasting and asset-risk group of eight. Returned at three days a week and moved to full time in March 2024.
- Owns the day-ahead load forecast for 2.4 million meters; mean absolute percentage error fell from 4.8% to 2.9%, about $6.1m a year in avoided imbalance charges.
- Built the outage-risk model that ranks 41,000 distribution poles for inspection; crews now find a defect on one visit in three rather than one in eleven.
- Set the review standard for any model that touches billing: written assumptions, a fixed backtest window, and a named owner. Eight models cleared it in the first year.
- Rewrote the forecast backtest to respect the data available at prediction time, which revealed that two years of reported accuracy had been optimistic by 0.6 points.
- Mentors three analysts; two have since shipped production models of their own.
Planned career break, Career break · June 2022 – August 2023
Full-time family care.
- Completed Harvard Extension School coursework in causal inference and shipped two releases of an open-source forecasting package.
Data Scientist, Alderbrook Health Analytics · March 2018 – May 2022
- Built the readmission risk model used across nine hospitals; 30-day readmissions in the flagged cohort fell 18% once care management began working its daily list.
- Replaced a purchased scoring tool with an in-house gradient-boosted model, matching its accuracy at a twelfth of the cost and making every feature auditable.
- Ran the outreach experiment program: 31 tests in two years, four of which changed standing policy.
Analyst, then Data Scientist, Marlowe Insight Partners · July 2015 – February 2018
- Forecast quarterly demand for 14 consumer brands, cutting overstock 27% against the planner-driven method it replaced.
- Automated the client reporting pack, returning about 30 analyst hours a month across the team.
Projects
- Seasoncast — An open-source seasonal forecasting package for Python, built for demand and load series with holiday effects.
Education
- M.S. Statistics · Boston University · September 2013 – May 2015
- B.S. Mathematics · University of Massachusetts Amherst · September 2009 – May 2013
Skills
- Modeling: Time series forecasting, Gradient boosting, Causal inference, Survival analysis, Hierarchical models
- Tools: Python, R, SQL, PyTorch, Spark, Airflow
- Practice: Backtesting, Model review, Experiment design, Executive briefings, Mentoring