Future algorithm adjustments

bvankat / nebrasketball

Tracking tournament odds for the Nebraska basketball team

http://nebrasketball.info

5 stars 1 forks source link

Future algorithm adjustments #11

Open bvankat opened 1 year ago

bvankat commented 1 year ago

Consider WAB as an additional resume metric?
- Does Torvik calculate this early in the season?
Bonus/penalty for Quad 1 wins and Quad 3/4 losses.
- 3-4 Q1 wins seems like the high end for 2022 bubble teams.
- Teams with 3 Q3 losses are very bubbly.
- Only one team in NET top 70 has more than 1 Q4 loss
Injury adjustments
- Could base on Sports Reference Win Shares (#)

bvankat commented 1 year ago

Consider: Adjust for victory/defeat margin within Q1/Q2 games. Nebraska didn't play many good teams close in 2023, lots of blowouts. The predictive metrics did not like that. SOR/KPI gave them credit for playing good teams, so there was a 50+ rank difference between the resume and predictive metrics.

UPDATE: Here's some detailed analysis of a similar erratic team: 2023 Missouri — https://www.rockmnation.com/2023/3/2/23621266/missouri-tigers-basketball-analysis-2022-23-net-rankings-explainer-ncaa-tournament-dennisgats

bvankat commented 1 year ago

Related to the margin comment above: Perhaps Kenpom Luck or Torvik FUN or Haslametrics Consistency metrics can be a proxy for explaining that variance? High-luck teams likely have wide gaps in metrics.

bvankat commented 6 months ago

Kenpom has good thoughts on how the quadrants should be broken down: https://theathletic.com/1465470/2019/12/17/kenpom-an-idea-of-a-better-quadrant-system-to-reward-bubble-teams/

Argues for smaller Quad 1
Eliminate the games against the lowest opponents.
Undefeated in Quad 3-4 is a sign of a good team

bvankat commented 6 months ago

Related to the margin comment above: Perhaps Kenpom Luck or Torvik FUN or Haslametrics Consistency metrics can be a proxy for explaining that variance? High-luck teams likely have wide gaps in metrics.

What about looking at margin in Quad 1 and 2 losses? Close losses indicate higher quality — especially road losses. Blowouts would maybe indicate they're less reliable/lower quality.

Nebraska's current losses in 2023-24 🫣:

Creighton by 26 (Home, Quad 1)
Minnesota by 11 (A, II)
Wisconsin by 16 (A, I)
Iowa by 18 (A, I)
Rutgers by 6 (A, II)
Maryland by 22 (A, II)

bvankat commented 6 months ago

Non-conference SOS should be a factor. Really poor ratings definitely affect how the committee views a team. Probably need some kind of penalty for a NCSOS above 250.