The Heart Disease Data Set. The Heart Disease data set is another example of a medical data set. In the data set results are stored of various medical tests carried out on a patient. The data set is also part of the statlog [76] data set repository. The data ≈ × set was constructed from a larger data set consisting of 303 records with 75 attributes each. For various reasons some records and most of the attributes were left out when the Heart Disease data set of 270 records and 13 input variables was constructed. The classification task consists of predicting the presence or absence of Heart Disease. The two target classes are quite evenly distributed with 56% of the patients (records) having no Heart Disease and 44% having some kind of Heart Disease present. The Heart Disease data set is quite small with only 270 records and 13 input variables. However, our simple gp still constructed 384 internal nodes as well as the 2 terminal nodes for the target classes, resulting in a search space of size 8.1 10105, which is larger than that for the German Credit data set. Table 2.7: Average misclassification rates (in %) with standard deviation, using 10-fold cross-validation for the Heart Disease data set. algorithm average s.d. simple gp 25.2 4.8 Ltree OC1 C4.5 15.5 29.3 21.1 478 cefr-miner esia 17.8 25.6 7.1 0.3 default 44.0 On this data set our simple gp algorithm performs significantly worse than the Ltree and cefr-miner algorithms. Compared to OC1, C4.5 and esia the differences in misclassification performance are not statistically sig- nificant.
The Heart Disease Data Set. The results on the Heart Disease data set are displayed in Table 3.8. All our new gp algorithms show a significant improvement in misclassification per- formance over our simple gp algorithm. Our best performing gp algorithm, gain ratio gp with k = 2, is significantly better than all our other gp algo- rithms, with the exception of the gain gp algorithm with k = 2. Based on the results it seems clear that two clusters or partitions per numerical valued attribute offer the best performance for our gp algorithms. With the exception of the gain gp algorithm using k = 3, all our new gp algorithms have a significantly lower misclassification rate than esia. The differences between our new gp algorithms and significant. cefr-miner are not The differences in classification performance between Ltree and our gp algorithms are statistically significant except for the two best performing gain ratio gp algorithms. OC1 performs very badly on this data set and as a result all our new gp algorithms are significantly better. The differences between C4.5 and our gp algorithms are not statistically significant.
The Heart Disease Data Set. The results on the Heart Disease dataset are displayed in Table 4.5. The differences between the fuzzy gp algorithms and their non-fuzzy versions are very similar and not statistically significant. All fuzzy gp algorithms perform significantly better than our simple gp algorithms, esia and OC1. Only the performance of our fuzzy gain ratio gp algorithm with k = 3 is not significantly worse than Ltree.
