Project Goal
- Use social media to develop additional informative inputs for an NBA draft prospect predictive model
- Determine whether a relationship exists between an NBA player’s social media presence and their on-court play
- Use that relationship to aid in draft prospect projections
Project Timeline
- Downloaded draft data (2010-2020)
- Scraped twitter
- Used IBM Natural Language Understanding API
- 3 sets of scores - Sentiment, Usage Rates, Topics
- XGBoost models predicting BPM
- BPM projection percentiles
- Dashboard
- Projections functionality
Modeling
- Trained on 80% of the draft data (2010-2020)
- XGBoost worked slightly more effectively than multiple regression
- Filled null BPM values (never debuted) at the 5th percentile
- Scores were not very predictive on their own
- Added a projection percentile to my analysis
| | RMSE | |
Sentiment | Usage | Topic |
---|
3.14 | 3.64 | 4.26 |
Next Steps
- Determine helpfulness of scores when used in other draft models
- Obtain all NBA player data
- Initially focused on current draft prospects
- Add similarity portion to projections tab