Inspiration
Coaches, scouts, and fantasy managers compare players using scattered box scores and gut feel. We wanted a fast, transparent way to see who's actually efficient, who's undervalued, and what a player's numbers mean for a lineup — without a black-box model nobody can explain.
What it does
GameIQ is a sports analytics dashboard built on real, per-game NBA box scores (2010-11 to 2023-24), aggregated into season averages for 6,100+ player-seasons.
- League Dashboard: season KPIs, top scorers, efficiency distribution, stat correlation matrix
- Player Comparison: two-player radar chart using league percentiles (so a guard and a center compare fairly), head-to-head bar chart, auto-generated insight
- Performance Trends: career trajectory across seasons for any player
- Hidden Gems: surfaces efficient, low-minute players (EFF per-36 minutes) — buy-low fantasy targets or players due for a bigger role
- MVP Predictor: transparent, user-adjustable weighted formula over real box-score stats (not a trained model) — validated against the actual 2023-24 MVP race
- Coach Recommendations: rule-based scouting report per player, generated from thresholds against league average
How we built it
Python, Pandas, and Plotly for the analytics and visualizations, Streamlit for the interface. Raw per-game box scores were filtered to games actually played, grouped into season averages, and used to derive advanced metrics: Efficiency Rating (EFF), True Shooting % (TS%), Effective FG% (eFG%), and Assist/Turnover ratio — all computed from real makes/attempts, not averaged percentages.
Challenges we ran into
Cleaning 14 seasons of raw box scores (DNPs, traded players, inconsistent minutes formatting) and turning raw counting stats into fair cross-position comparisons — solved with percentile-based radar scoring instead of raw stat overlays.
Accomplishments
A fully explainable analytics pipeline where every insight traces back to a real formula, not a black box — and an MVP formula that independently reproduces the real 2023-24 MVP top 5.
What we learned
How much a good aggregation layer (season averages, shooting splits from summed makes/attempts) matters before any "insight" can be trusted.
What's next
Injury-risk overlay using workload data, a team-level roster fit optimizer, and a clutch-performance module using play-by-play data.
Log in or sign up for Devpost to join the conversation.