Subgroup visualization is very valuable for expert interpretation of subgroup discovery results. From Figures 6 and 7 it can be seen that there is no significant difference between CHD patients and healthy subjects regarding their age, but that there are significant differences among the detected patterns. Figure 8 illustrates a similar effect for the total cholesterol values although it is known that total cholesterol is an important risk factor for the CHD disease. This observation shows that the problem of CHD risk group detection can typically not be solved by considering single features and demonstrates the appropriateness of the suggested approach which tries to generate subgroup descriptions which are a logical conjunction of a few correlated features.
Figure 10 is also interesting, since it is very different from other figures. Notice that exercise ECG ST segment depression was not used as an attribute in the training data (which contained only attributes that are available at stages A, B and C); exercise ECG ST segment depression, long term ECG recording and echochardiography are not available for early risk group detection since they can be collected/measured only in specialized medical institutions. Figure 10 clearly demonstrates significant differences between all CHD and all healthy subjects in terms of exercise ECG ST segment depression values, demonstrating that this measurement, if available, is an excellent disease indicator. But it also shows that, although it is known that patterns A1 and C1 cover different disease subpopulations, they behave very similarly in terms of the exercise ECG ST segment depression property.