A Decision Tree Approach to Predicting Business Failure

1
A Decision Tree Approach to Predicting Business
Failure

• Sam Hallman
• CS175 6/4/09

2
What is a decision tree?
3
The data
4
80 attributes per record
5
80 attributes per record(20 attributes per
quarter)
6
Approach 1

• attribute11/07, attribute21/07, ,
  attribute201/07,attribute14/07, attribute24/07,
  , attribute204/07, attribute17/07,
  attribute27/07, , attribute207/07,
  attribute110/07, attribute210/07, ,
  attribute2010/07

7
(What the results look like)

• CTD038_10_07 gt 0 Bad (19487.0/3758.3)
• CTD038_10_07 lt 0
• LGC004_10_07 gt 0 Bad (12223.0/2588.8)
• LGC004_10_07 lt 0
• CLC006_7_07 gt 0 Bad (1766.0/603.0)
• CLC006_7_07 lt 0
• CTD038_4_07 gt 0 Bad (5155.0/1956.6)
• CTD038_4_07 lt 0
• IQC003_7_07 gt 0 Bad
  (2092.0/861.3)
• IQC003_7_07 lt 0
• CTD038_1_07 gt 0 Bad
  (4200.0/1800.5)
• CTD038_1_07 lt 0
• LGC004_7_07 gt 0 Bad
  (93.0/37.4)
• LGC004_7_07 lt 0
• CTB019_10_07 lt 800
• RTO077_4_07 lt 3
  
• JDB004_1_07 gt
  2 Bad (104.0/47.8)
• JDB004_1_07
  lt 2
• 
  RTO077_10_07 lt 3

8
Approach 2
9

• avg1,avg2,avg3,avg4,avg5,avg6,,avg19,avg20avg1,a
  vg2,avg3,avg4,avg5,avg6,,avg19,avg20avg1,avg2,av
  g3,avg4,avg5,avg6,,avg19,avg20avg1,avg2,avg3,avg
  4,avg5,avg6,,avg19,avg20avg1,avg2,avg3,avg4,avg5
  ,avg6,,avg19,avg20avg1,avg2,avg3,avg4,avg5,avg6,
  ,avg19,avg20avg1,avg2,avg3,avg4,avg5,avg6,,avg1
  9,avg20.

10
Approach 3
11
Approach 3 - Illustration
12
Approach 4
13
Percentage error on testing un-pruned trees on
the test set
14
Problem
15
Matthews Correlation Coefficient
16
(No Transcript)