The German Credit Data Set. The German Credit data set also comes from the statlog data set repos- itory [76]. The original data set consisted of a combination of symbolic and numerical attributes, but we used the version consisting of only numerical valued attributes. The data set is the largest data set used in our experiments with 1000 records of 24 attributes each. The two target classes are divided into 700 examples for class 1 and 300 examples for class 2. Although the data set itself is the largest one we used, the simple gp only constructed 269 possible internal nodes as well as 2 terminal nodes. As a result the search space of our simple atomic gp on the German Credit data set is much smaller (size ≈ 1.3 × 10101) than on the Australian Credit data set.
The German Credit Data Set. When we consider the results on the German Credit data set in Table 3.6 we see that both our simple gp and gain gp algorithm with k = 3 perform significantly better than our other gp algorithms, with the exception of our clustering gp algorithm with k = 2. The differences with the decision tree algorithms Ltree, OC1 and C4.5 are not statistically significant. If we compare the results of our gp algorithms with esia we see that the differences between our simple gp and clustering gp algorithms and esia are statistically significant. In the case of our gain gp algoritms the differences with esia are also statistically significant except for k = 4. Differences be- tween esia and our gain ratio gp algorithms are not statistically significant.
