Data Scientist
2021 - 2022Visimo- Built a graph neural network to predict risk on a construction site. It reduced error relative to the prior model by 60% and provided reasonable results when tested on individual observations that differed from training.
- Delivered a resource management tool that simulated project wins and losses and assessed the risk of understaffing. R Shiny was used to give users an interface to run the process and explore simulation results.
- Conducted data discovery for a healthcare client to assess the state of their data and explore insights that could be extracted with deep learning for drug development. This led to a monthly retainer and future work as data came in.
Technologies: Julia, R, Python, Docker, Neo4j, Data Visualization, Simulations, Deep Learning, Predictive Modeling, Statistics, SQL, Machine Learning, Data Science, RStudio, RStudio Shiny, Healthcare IT, Artificial Intelligence (AI), Linear Optimization, Neural Networks, Pandas, Clustering, Data Analysis, Data AnalyticsSenior Actuarial Consultant
2018 - 2021Talcott Resolution- Overhauled four major assumptions to use a predictive modeling framework that could efficiently handle the millions of records on policyholder behavior that went into premises and was much more accurate than the prior approach. Used R and H2O.
- Replaced a compression process (clustering to reduce time with Monte Carlo simulations) with a modern custom solution that reduced cloud costs by 1/3 and improved the accuracy of the simulation.
- Built an investment management tool that calculated the efficient frontier for investments given unique life insurance portfolio constraints.
- Investigated optimal mapping of mutual funds to relevant major indices and built a process to identify which holdings were responsible for inaccuracies in that mapping.
- Implemented a novel method of setting prudency in assumptions to reflect the appropriate level of conservatism when setting reserves.
- Assisted assumption reviews for several M&A projects.
- Built efficient reporting frameworks for financial simulations, policyholder behavior, and call center data.
Technologies: R, Excel VBA, Simulations, Data Visualization, Predictive Modeling, Life Insurance, Financial Mathematics, Tail Risk, Statistics, SQL, Machine Learning, Data Science, RStudio, RStudio Shiny, Artificial Intelligence (AI), Linear Optimization, H2O Deep Learning Platform, Clustering, Data Analysis, Data Analytics
Daniel Doyle
Data Scientist and Developer in Pittsburgh, PA, United States
Member since May 3, 2022
Daniel is a data scientist who builds predictive models, data visualizations, and dashboards on large datasets with expertise in the staffing and investment management fields. He made a graph neural network to predict risk on a construction site reducing errors by 60%. Daniel improved a predictive modeling framework, handling millions of records on policyholder behavior using R and H2O. His models are used in financial forecasts and dashboards to catch aberrant behavior as it unfolds.
Daniel is now available for hire
Portfolio
- VisimoJulia, R, Python, Docker, Neo4j, Data Visualization, Simulations...
- Talcott ResolutionR, Excel VBA, Simulations, Data Visualization, Predictive Modeling...
Experience
- R 4 years
- Simulations 4 years
- Machine Learning 4 years
- Statistics 4 years
- Financial Mathematics 3 years
- Predictive Modeling 3 years
- Python 1 year
- Julia 1 year
Availability
Part-time
Preferred Environment
Docker, Julia, R, Python
The most amazing...
...project I have completed is predicting construction site risk. It used custom text embeddings and GNNs to greatly improve performance.
Employment
Experience
- Personal Investment Managementhttps://github.com/DDoyle1066/InvestmentManagement
This process generates optimally diversified portfolios based on different risk thresholds. It proceeds by:
1. Gathering returns from AlphaVantage based on requested tickers (currently Vanguard ETFs and a high yield mutual fund).
2. Estimating the relationship between bond yields and bond funds.
3. Generating a process to project variations in bond yields and ETF returns if bond yields are relevant.
4. Simulating future bond yields and ETF movements based on them.
5. Determining the efficient frontier for that simulation.
6. Repeating #4 and #5 1,000 times to avoid overallocation into a single high-performing fund at the end of the efficient frontier. - Variable Annuity Metamodelinghttps://github.com/DDoyle1066/VA_Metamodeling
This project walks through pricing a block of variable annuities and developing a faster metamodel for that approximation. It proceeds by:
1. Generating policyholder in force files.
2. Pulling historical market data and estimating market parameters from historical data. A regime-switching lognormal model is shown as an improvement over a simple lognormal model.
3. Monte Carlo calculation.
4. Approximation of projection using a neural network to predict the output without running the full-scale Monte Carlo projection.
Skills
Languages
Julia, R, Excel VBA, Python, SQLFrameworks
RStudio ShinyParadigms
Data SciencePlatforms
RStudio, H2O Deep Learning Platform, Docker, Jupyter NotebookOther
Life Insurance, Predictive Modeling, Probability Theory, Simulations, Artificial Intelligence (AI), Linear Optimization, Data Analysis, Data Analytics, Statistics, Risk Models, Tail Risk, Financial Modeling, Financial Mathematics, Data Visualization, Deep Learning, Machine Learning, Neural Networks, Clustering, Bayesian Statistics, Healthcare IT, Principal Component Analysis (PCA), Multivariate Statistical Modeling, Linear Regression, Monte Carlo SimulationsLibraries/APIs
Pandas, TensorFlow, NumPy, MatplotlibTools
Git, Jupyter, SeabornStorage
Neo4j
Education
- Bachelor's Degree in Actuarial Science2015 - 2018Robert Morris University - Moon Township, Pennsylvania, United States
Certifications
- Associate of the Society of ActuariesJANUARY 2020 - PRESENTSociety of Actuaries
