Further runs have confirmed the kind of distribution in Table 1 of the paper. In addition, quite a few "ten times ten-fold crossvalidations" have been done with the result that under somewhat modified parameters, the mean test score is in the neighborhood of .961. I would guess that if you are getting results that vary significantly more than what is shown in Table 1, there is some difference in your program causing it. But real statistics haven't been done as yet. I suppose that's because the results so far look good compared with results from other techniques, plus insufficient knowledge of proper statistical methods. Most of the effort up to now has been on parameter variations and this gave the higher score just mentioned. (12/17/00).
I am attempting to reproduce your results with XCSI applied to the Wisconsin Breast Cancer Database. I have had some success to date but to progress further I could do with some assistance.
Do you have any statistics on the variability of results obtained from any additional data produced from your experiments in order to evalutate my program's performance?