南航暑期国际课程大数据可视化第4讲2.ppt

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * VIMONO demonstration for BREAST CANCER DIAGNISTICS Figure 7. Visualization of expert cancer and biopsy rules VOMONO for Expert cancer and biopsy rules Figure 7 shows the advantages of chain-based MDF visualization relative to visualization that does not exploit monotone chains. The chain-based border between classes is much clearer. This advantage gives an immediate benefit: visual comparison of rules for biopsy and cancer. It also helps to identify the level of consistency of cancer and biopsy rules provided by the expert. It is expected that a biopsy rules should be less stringent than a cancer rules. For instance if the presence of x2x3 is a cancer indicator but only presence of x3 can be sufficient to recommend biopsy test. In visual terms it means that the border of the biopsy pattern should be lower or at the same as cancer pattern for them to be consistent. This is exactly the case as Figure 7 (c) and (f) show. The black areas in the ovals in Figure 7 (f) for biopsy are lower than the same areas for cancer in Figure 7 (f). Cancer Biopsy 4. GENERAL CONCEPT OF USING MDF IN DATA MINING Figure 8 illustrates the general concept of simultaneous coordinated visualization of multiple components of the analytics. This figure shows that rules “extracted” from the expert in this example are coordinated and aligned with each other, the data and the visual pattern better than the rule extracted from data by analytical data mining algorithm. Comparison of final and warning rules Comparison of rules and data SCALING ALGORITHMS: Algorithm with data-based chains Our goal is to represent the MDF on a single screen with possibly some scrolling. The major factor that is limiting the visualization is the number of Hansel chains that can be visualized as vertical lines. We use two approaches: (A1) grouping chains with similar height of the border to a cluster, (A2) constructing chains from only