HammingNN classifier: results with various datasets

2012-12-19

This table presents results of using the HammingNN classifier with a variety of publicly available datasets that are frequently used for development and testing of classifier paradigms. For comparison purposes, I’ve included results from 3 other classifier paradigms.

Note: you may need to widen your browser window to view all the columns.

Data Set Cases Attributes Classes HammingNN
leave-one-out
HammingNN
10-fold
Parameters XCSTS C4.5  Naive Bayes
Anneal 898 32D 6C 5 99.67% 99.496% n=21 i=28 k=2 97.7% 98.6 86.6
Breast cancer Wisconsin 699 9D 2 96.567% 96.383% i=8 j=3 95.9 94.5 96.0
BUPA 345 6C 2 70.72% 69.67% n=15 67.1 65.0 54.8
Glass 214 9C 7 76.64% 75.45% g n=19 i=5 71.8 67.4 48.1
Ionosphere 351 32C 2 93.447% 92.829% n=12 i=23 90.1 90.0 82.5
Iris 150 4C 3 98.000% 97.960% i=2 n=3 94.7 94.8 95.5
Leukemia 72 5147C 2 97.222% 95.028% n=8 i=11
Mushroom 8124 22D 2 100% 100% i=7 100.0 100.0 95.8
Prostata 102 12533C 2 94.118% 92.451% n=6 i=3
Soybean 683 35D 19 94.290% 93.634% z i=25 85.1 91.9 92.8
Vehicle 846 18C 4 70.567% 70.051% i=18 n=13 74.1 72.4 45.1
Vowel 990 10C 11 98.687% 98.087% g n=13 66.0 80.1 63.2
Wine 178 13C 3 97.191% 96.708% n=5 95.6 93.3 97.2
Yeast 1484 8C 10 58.154% 57.023% u=5 i=6
Zoo 101 16D 7 98.02% 97.911% i=12 95.1 93.0 95.7

Data Sets: from the UIC Machine Learning Repository; I used the versions of these datasets downloaded from the Orange website: http://orange.biolab.si/datasets.psp

Attributes: D = discrete (ie nominal or ordinal or categorical); C = continuous-valued.

Parameters:

  • i = number of attributes used;
  • n = number of “slices” for continuous attributes;
  • u = upper limit for number of slices (when number of slices is automatically calculated);
  • j, k, z, g: parameters specifying, respectively, depth value, variable margin, one bit per class for discrete attributes, graded bits for continuous attributes.

The results in the XCSTS, C4.5, and Naive Bayes columns are from Table 9.3 in the book: Butz MV. Rule-based evolutionary online learning systems : a principled approach to LCS analysis and design. Berlin: Springer; 2006:xxi, 266.

Leave a Reply

Your email address will not be published.

This site uses Akismet to reduce spam. Learn how your comment data is processed.