NBA FINALS WHO

1
NBA FINALS WHO?

• A Neural Networks Approach
• By
• Asong Tambo

2
OVERVIEW

• Project objective
• Reasons
• Network design
• Data collection methods
• Analysis
• Sample results
• Explanations
• Conclusion

3
The Objective

• Win or go home!
• Use a neural network to determine which team goes
  to the finals

4
REASONS

• No one has done this before (I think)
• No more heartache!

5
NETWORK DESIGN

• Positions on the court
• Initial network units
• 4 inputs
• 1 hidden layer (2 units)
• 1 output

6
DATA COLLECTION

• www.nba.com
• From 1997 2004 (8yrs)
• 28 of the 29 NBA teams were functional
• Potentially 196 entries
• Key questions
• Are there KEY players on this team?
• How many players contribute to the success of the
  team?

7
DATA COLLECTION

• Definitions
• Key player one who satisfies any 2 of the
  following 3 conditions
  
• 20 PPG
• 5 APG
• 5 SPG
• Team average/contribution
• Initially, 5 or more players with 20 PPG

8
ANALYSIS

• First few runs were absolutely pathetic!
• Could not distinguish between and - examples

9
ANALYSIS

• Tried several idea
• Actual values for team average instead of (1,0)
• Increase number of nodes in hidden layer (2 -
  4)
  
• Normalize team average between (1,0)
• There are 2 Guards and 2 Forwards

10
ANALYSIS
11
ANALYSIS

• EUREKA!!!
• 86 training examples, lots of cycles
• Network recognized some of the positive examples
  and was close on others.
  
• 87 examplestotal error 5.548error/ex 0.0638

12
SAMPLE DATA - negative

• Center 0
• Forward 0
• Guard 0
• Team avg .26
• Target 1
• Output 0.22

13
SAMPLE DATA - positive

• Center 1
• Forward .5
• Guard 0
• Team avg .63
• Target 1
• Output 0.977

14
SAMPLE DATA - positive

• Center 0
• Forward 0
• Guard 0
• Team avg .8
• Target 1
• Output 0.997

15
CONCLUSION

• High hi probability of appearing in
  finals
  
• Compare for the 2 positive cases
• Further testing needed

16
QUESTIONS?