Separability of Split Value based tree results

Dataset
Whole set 
accuracy / error
LOO
(best first)
LOO
(beam)
  10-fold CV   (best first)
min / avg / max
10-fold CV (beam)
min / avg / max
Appendicitis
94.34 / 5.66
12.26
10.38
12.18 / 13.86 / 16.86
11.12 / 13.72 / 14.86
Cleveland Heart
85.81 / 14.19
19.14
  
17.84 / 20.55 / 22.40
18.19 / 19.82
Ljubliana breast cancer
77.62 / 22.38
    
26.88 / 27.87 / 29.73
25.51 / 26.46 / 27.24
Wisconsin breast cancer
97.42 / 2.58
3.86
  
4.57 / 4.92 / 5.44
3.43 / 3.68 / 3.87
Iris
98.00 / 2.00
        
Mushroom
100.00% / 0.00%
0.00
0.00
0.00
0.00

Datasets with training and test parts

Dataset
Training data
(best first)
Test data
(beam)
Training data
(best first)
Test data
(beam)
Monks 1
100% / 0%
93.06% / 6.94%
100% / 0%
100% / 0%
Monks 2
 100% / 0%
80.56% / 19.44%
 
  
Monks 3
93.44% / 6.56%
97.22% / 2.78%
93.44% / 6.56% 
 97.22% / 2.78%
Hypothyroid
100% / 0%
99.15% / 0.85%
99.79% / 0.21%
99.33% / 0.67%
Shuttle
100% / 0%
99.99% / 0.01%
100% / 0%
99.99% / 0.01%

Rule sets (for the whole datasets)

Appendicitis

Accuracy = 94.34%, error = 5.66%

  1. HNEA < 7520.5 AND MBAP < 12 -> class 0
  2. HNEA < 9997.5 AND HNEA > 9543.5 -> class 0
  3. else -> class 1

Cleveland Heart

Accuracy = 85.81%, error = 14.19%

  1. ca = 0.0  AND exang = 0  -> class 1
  2. NOT cp = 2  AND  NOT slope = 2  -> class 1
  3. ca = 0.0  AND  AND thal = 0  -> class 1
  4. else -> class 2

Ljubliana breast cancer

Accuracy = 76.22%, error = 23.78%

  1. inv-nodes > 2 AND deg-malig in [2,4] -> class recurrence-events
  2. else -> class no-recurrence-events

Accuracy = 77.62%, error = 22.38%

  1. (tumor-size = 25-29  OR tumor-size = 30-34  OR tumor-size = 35-39  OR tumor-size = 45-49 ) AND deg-malig < 2.5 -> class no-recurrence-events
  2. NOT (tumor-size = 25-29  OR tumor-size = 30-34  OR tumor-size = 35-39  OR tumor-size = 45-49 ) AND node-caps = no  -> class no-recurrence-events
  3. (tumor-size = 25-29  OR tumor-size = 30-34  OR tumor-size = 35-39  OR tumor-size = 45-49 ) AND  NOT menopause = premeno  AND irradiat = no  AND  NOT tumor-size = 30-34  -> class no-recurrence-events
  4. else -> class recurrence-events

Wisconsin breast cancer

    Accuracy = 97.42%, error = 2.58%
    1. F4 > 2.5 AND F7 > 2.5 -> class 2
    2. F4 > 2.5 AND F7 < 2.5 AND F6 > 3.5 -> class 2
    3. F4 < 2.5 AND F7 > 1.6734235 AND F2 > 5.5 -> class 2
    4. else -> class 1

Iris

Accuracy = 98.00%, error = 2.00%

  1. F4 < 0.8 -> class 1
  2. F4 > 1.65 -> class 3
  3. F3 > 4.95 -> class 3
  4. F4 > 0.8 AND F4 < 1.65 AND F3 < 4.95 -> class 2

Mushroom

Accuracy = 100%, error = 0%

  1. odor != a AND odor != l AND odor != n -> class 1
  2. spore_print_color = r  -> class 1
  3. gill_size = n AND (stalk_surface_above_ring = y OR stalk_surface_above_ring = k OR population = c) -> class 1
  4. else -> class 0

    5.  

Hypothyroid

Training: Accuracy = 99.79%, error = 0.21%. Test: Accuracy = 99.33%, error = 0.67%

  1. thyroid_surgery = 0 AND TSH > 0.00605 AND FTI < 0.06472 -> class 1
  2. thyroid_surgery = 0 AND TSH > 0.00605 AND FTI > 0.06472 AND on_thyroxine = 0  AND TT4 < 0.1505 -> class 2
  3. else -> class 3