Visual Data Mining
4 likes2,010 views
This document discusses visual data mining using HeatMiner software. It shows how increasing amounts of sample data (from 10 to 100,000 samples) shapes visually into patterns that reveal relationships between a value of interest and other attributes. With only 10,000 samples, the data forms three distinct "donuts", and adding more samples simply reduces noise without changing the overall patterns or conclusions that can be drawn from the data.
Visual Data Mining
- 1. Visual Data Mining HeatMiner® is a registered trademark of Agience Oy Ltd by Pauli Misikangas, the CEO of Cloud’N’Sci Ltd
- 2. Mysterious Data The Value Possibly Related of Interest Attributes How does the Value of Interest depend on X,Y,Z ? Algorithms-as-a-Service cloudnsci.fi
- 3. 10 Samples Colors indicate the Value of Interest (xyz+color = 4D visualization) Samples separated, Attributes no visual patterns X,Y and Z → more data needed define a 3D space Algorithms-as-a-Service cloudnsci.fi
- 4. 100 Samples Adjacent samples merge and form a ’heatmap’ in which colors indicate the average value Algorithms-as-a-Service cloudnsci.fi
- 5. 1000 Samples The secrets of the data are shaping visually... Algorithms-as-a-Service cloudnsci.fi
- 6. 10000 Samples Now it’s easy to see that data points form three ’donuts’! HeatMiner® is a registered trademark of Agience Oy Ltd Algorithms-as-a-Service cloudnsci.fi
- 7. 100000 Samples Further samples reduce ’bumpiness’ of the 4D heatmap The visual illusion and conclusions are still the same – no more data needed Algorithms-as-a-Service cloudnsci.fi
- 8. Visual Data Mining RIGHT !?! CONCLUSIONS & ACTIONS AUTOMATIC Raw input DATA REFINING Visual illusion data ALGORITHMS of the data Algorithms-as-a-Service cloudnsci.fi
- 9. Questions? Comments? Ideas? Use Cases? Algorithms-as-a-Service cloudnsci.fi