Dear forum members,
Having undertaking a large experimental series usinf the J48 classifier in Weka 3.6.0 experimenter I hit one unusual result.
Runningone of my (unpublished as yet) datasets with confidence intervals of 0.25, 0.2, 0.15, 0.1 and 0.05 th baseline CA (conf = 0.25) of 88.73 was obtained. The next value of 89.31 (conf = 0.2) was flagged (with v) as statistically larger. Successive values of 89.23 (conf=0.15), 89.35 (conf=0.1) and 89.37 (conf=0.05) were not flagged as significantly different from the baseline.
As the difference is very small (0.58) compared with other statistically significant differences from my experiments I am led to the belief that this is simply a reporting error in Weka - has anyone else had a similar experience, or any suggestions to explain the behaviour?
Thanks,
Jim
