Looking for Data Scientist for Predict College Basketball Game Outcomes
Budget: $1,500 – $3,000 USD
I want to turn my rich collection of college-basketball data into a model that reliably forecasts game results before tip-off and updates the probability as plays unfold. You’ll be working with three data streams I already have in place—historical game logs, detailed player performance metrics, and a live in-game feed.
The core of the job is to design, train, and validate a predictive pipeline in Python that combines Monte Carlo simulation with machine-learning techniques. I’m especially interested in seeing how simulation at both the game level and the play-by-play level can blend with gradient-boosting, neural networks, or any other supervised approaches you believe outperform a pure simulation baseline. Robust statistical testing and model-evaluation routines are essential so we can quantify lift over publicly available benchmarks.
Because the ultimate goal is dependable win-probability estimates, I need clear deliverables:
• Clean, well-documented Python code (Jupyter notebooks or .py modules)
• Reproducible training scripts tied to my data folders
• Version-controlled experiments showing validation accuracy and calibration metrics
• A short technical report summarizing methodology, assumptions, and recommendations for next steps
If you’ve previously built Monte Carlo or machine-learning models for sports—especially basketball—tell me about the algorithms you used, how you handled lineup changes, and any performance gains you achieved. I’m ready to iterate quickly, provide domain guidance.
The core of the job is to design, train, and validate a predictive pipeline in Python that combines Monte Carlo simulation with machine-learning techniques. I’m especially interested in seeing how simulation at both the game level and the play-by-play level can blend with gradient-boosting, neural networks, or any other supervised approaches you believe outperform a pure simulation baseline. Robust statistical testing and model-evaluation routines are essential so we can quantify lift over publicly available benchmarks.
Because the ultimate goal is dependable win-probability estimates, I need clear deliverables:
• Clean, well-documented Python code (Jupyter notebooks or .py modules)
• Reproducible training scripts tied to my data folders
• Version-controlled experiments showing validation accuracy and calibration metrics
• A short technical report summarizing methodology, assumptions, and recommendations for next steps
If you’ve previously built Monte Carlo or machine-learning models for sports—especially basketball—tell me about the algorithms you used, how you handled lineup changes, and any performance gains you achieved. I’m ready to iterate quickly, provide domain guidance.