Information Technology Reference
In-Depth Information
Table 3.2.
comparison of speed convergence
-
AdaBoost M1
BrownBoost
AdaBoost hyb
Nb. iterations 10 20 100 1000 10 20 100 1000 10 20 100 1000
Iris 7,00 6,00 5,90 5,85 3.96 3.89 3,80 3,77 3,50 3,00 3,00 3,00
Nhl 37,00 35,00 34,87 34,55 30,67 30,01 29,89 29,76 31,00 28,00 28,00 28,00
Weather 21,50 21,42 21,40 14,40 21,10 21,00 20,98 21,95 21,03 21,00 21,00 21,00
Credit-A 15,85 15,79 15,75 14,71 13,06 13,00 12,99 12,97 14,00 13,91 13,91 13,91
Titanic 21,00 21,00 21,00 21,00 24,08 24,00 23,89 23,79 21,00 21,00 21,00 21,00
Diabetes 27,70 27,61 27,55 27,54 25,09 25,05 25,03 25,00 25,56 25,56 25,56 25,56
Hypothyroid 0,60 0,51 0,51 0,50 0,62 0,60 0,59 0,55 0,43 0,42 0,42 0,42
Hepatitis 16,12 15,60 14,83 14,19 14,15 14,10 14,08 14,04 14,03 14,00 14,00 14,00
Contact-Lenses 26,30 24,80 24,50 16,33 15,90 15,86 15,83 15,80 16,00 16,00 16,00 16,00
Zoo 7,06 7,00 7,00 7,00 7,25 7,23 7,19 7,15 7,00 6,98 7,00 7,00
Straight 2,50 2,46 2,45 2,42 2,12 2,00 1,98 1,96 0,42 0,42 0,42 0,42
IDS 2,00 1,90 1,88 1,85 0,7 0,67 0,65 0,63 0,7 0,67 0,65 0,63
Lymph 19,53 19,51 19,51 19,50 18,76 18,54 18,50 18,45 18,76 18,54 18,50 18,45
Breast-Cancer 45,89 45,81 45,81 45,79 31,10 31,06 31,04 31,00 31,10 31,06 31,04 31,00
Convergence speed
10 20 100 1000
Number of iteration
Fig. 3.4.
Average Error by iteration
For this reason we choose for the first part 20 iterations to carry out the
comparison in terms of error and recall. These results are also valid for the
database Hepatitis. In fact, This database has a lot of missing values (Rate 6%).
These missing values always present a problem of convergence. Moreover, the
same results appear on databases of various types (several attributes, the class
to be predicted with K modalities, important sizes).
