SqO — NBA Shot Quality & Optimization

A spatial analytics tool that generates personalized shot quality maps for individual NBA players using Gaussian Kernel Density Estimation (KDE) on real shot chart data from the NBA Stats API.

Built as a final project for an AI/ML course at the University of Connecticut.

The Problem

League-average shot quality models treat all players the same. But we know that is not true, efficiency relies on context and that difference is the entire point of this project.

What It Does

• Pulls real shot chart data via nba_api (NBA Stats API)
• Builds a continuous Points Per Shot (PPS) surface using Gaussian KDE with a Bayesian prior for sparse zones
• Generates player shot quality heatmaps normalized within each scoring zone
• Produces a zone-level EV audit table across 4 strategic zones
• Outputs a defensive scouting report with scheme recommendations

Key Findings — 2023-24 Season

Player	Rim EV	Corner 3 EV	Midrange EV	Above-Break 3 EV
Stephen Curry	1.244	1.571 (+41% vs avg)	0.830	1.183
Giannis Antetokounmpo	1.472	0.000	0.690	0.850
DeMar DeRozan	1.227	1.167	0.916 (+17% vs avg)	0.894
League Average	~1.20	~1.11	~0.78	~1.05

Shot Quality Signatures

Methodology

• Fully deterministic — same data = same output every run
• Statistically principled for 1,000–1,500 spatial point events
• Runs in <60 seconds on CPU, no GPU needed
• More interpretable to coaching stakeholders than a black-box model

Bayesian Prior: 5 pseudo-shots at league average FG% (0.46) anchors estimates in sparse zones without distorting data-rich zones.

Within-Zone Normalization: Color is scaled independently inside and outside the arc, revealing subtle efficiency gradients hidden by the 2pt vs 3pt value difference.

Tech Stack

Python · nba_api · NumPy · SciPy · Pandas · Matplotlib

Setup

pip install nba_api scipy matplotlib numpy pandas

Then run SQO.ipynb in Google Colab or Jupyter.

Players Covered

The notebook runs on 5 players by default:

• Stephen Curry (Elite Shooter)
• Giannis Antetokounmpo (Rim Attacker)
• DeMar DeRozan (Midrange Master)
• Kevin Durant (Versatile Scorer)
• Nikola Jokic (Post Scorer)

To add any player, find their NBA Stats ID and add to the PLAYERS list.

Roadmap

• Defender proximity conditioning (open vs contested surfaces)
• Game-state context (shot clock, score differential, fatigue)
• Opponent scouting module (defensive vulnerability maps)
• Real-time REST API deployment

Research Paper

Full write-up in /paper/SQO_Paper.pdf covering methodology, related work (Goldsberry 2012, Cervone et al. 2014), and findings.

References

1. Goldsberry, K. (2012). CourtVision. MIT Sloan Sports Analytics Conference.
2. Cervone et al. (2014). POINTWISE. MIT Sloan Sports Analytics Conference.
3. Franks et al. (2015). Counterpoints. MIT Sloan Sports Analytics Conference.
4. Diggle, P.J. (2013). Statistical Analysis of Spatial Point Patterns. CRC Press.
5. nba_api