Basketball Lineup Analytics Engine

A matchup-aware lineup optimization and rotation decision-support tool built for basketball analytics.

This application evaluates every viable 5-man lineup combination using advanced player metrics, contextual game-state adjustments, and a supervised machine learning model to project net rating impact.

Designed as a sports analytics / basketball operations tool.

👉 Live Demo: (https://basketball-lineup-analytics-engine.streamlit.app/)

What this does

Lineup Recommendations

• Ranks all 5-man combinations from a minutes-based player pool
• Supports roster constraints (min guards, min bigs, max centers)
• Optimizes for game-state priorities (Balanced / Need a bucket / Get stops / Protect the ball / Win the glass)

Substitution Assistant

• Given the current 5 + bench, suggests the best one-for-one substitutions
• Uses the same objective + constraints as the lineup ranking engine

Matchups vs ACC Opponents

• Scores Syracuse 5-man counters vs a specific opponent 5-man lineup
• Opponent lineup modes: auto (top minutes), archetype (Big/Small/Shooting/Defense), manual

Screenshots

Lineup Recommendations

Substitution Assistant

Matchup vs Opponent

Modeling Approach

• Player-level season metrics aggregated into lineup-level features
• Supervised regression model predicts projected net rating
• Reliability penalty applied for low-minute units
• Fatigue multiplier adjusts impact late-game
• Chemistry & familiarity multipliers stabilize predictions
• Matchup adjustment estimates opponent interaction effects
• Final Objective score = weighted normalized composite

This system is built as a decision-support engine — not a deterministic predictor.

Tech Stack

• Python
• Streamlit
• Pandas / NumPy
• Scikit-Learn
• Joblib

Data inputs

• data/syracuse_player_profiles.csv (Syracuse player season profiles)
• data/acc_player_profiles.csv (ACC player season profiles)
• data/lineup_model.joblib (trained model used to predict lineup net rating)

Run locally

git clone https://github.com/jacobsmith0408-svg/basketball-lineup-analytics-engine.git
cd basketball-lineup-analytics-engine

python -m venv .venv
source .venv/bin/activate   # Windows: .venv\Scripts\activate
pip install -r requirements.txt

streamlit run app.py

Project Structure

• app.py → Streamlit interface + scoring engine
• src/ → data processing / model scripts
• data/ → input CSVs + trained model artifact

Limitations

• Model performance depends on feature quality and training data
• Matchup logic is heuristic-based (Level-1 interaction layer)
• Designed as a prototype decision-support tool, not a production analytics platform

Engineering Notes

• Deployed via Streamlit Cloud with GitHub CI-based redeploys.
• Path-safe file loading using pathlib to support Linux-based cloud environments.
• Constraint-based lineup generation using combinatorics (itertools.combinations).
• Model predictions layered with reliability penalties, fatigue multipliers, and matchup heuristics.
• Results exportable to CSV for scouting workflow integration.