Fig. 1. The conceptual framework of the mixed-initiative synthesized learning approach.

Fig. 2. “Competitive Learning” Concept in the SOM.

Fig. 3. The concept of the distance measure of the automatic labeling technique.

Fig. 4. The architecture of the mixed-initiative synthesized learning approach.

Fig. 5. The upper-left window shows the boundaries of clusters and the number of clusters; the bottom-right window shows the ranges (Maximum, Minimum, Average) for each labeled feature of a particular cluster.

Fig. 6. The interface for cluster class assignment.

Fig. 7. The conceptual diagram of cross-class analysis.

Fig. 8. CRMiner sysem.

Fig. 9. The interface for users’ attribute selection.

Fig. 10. The results of decision tree learning in terms of a set of class characteristic rules.

Fig. 11. The screen dump showing a portion of the results obtained after cross-class analysis.

Fig. 12. Three different factors that can be controlled by users.

Fig. 13. The relationship between the attribute number and the cluster number/rule number in the eight experiments.

Fig. 14. The relationship between the attribute number and averaged attribute differential ration in the eight experiments.

Fig. 15. The relationship between a and the cluster number in the experiments.

Fig. 16. The relationship betweenaand the averaged attribute differential ratios.

Fig. 17. The averaged attribute differential ratios for the 10 experiments conducted by a same human analyst.

Table 1. Integrated attribute names and their meaning

Table 2. The final tabulated cross-class analysis result with α=9

Table 3. The tabulated results for different settings of selected attributes/attribute number

Table 4. The tabulated results of experiments with different values of α

Table 5. The tabulated results of 10 experiments conducted by a human analyst