Interactive Comprehensible Data Mining - Ambient Intelligence for Scientific Discovery - page 49

1.2 Key Concepts

The key concepts in the Haiku system are

- Interaction via visual representation of data - visualization used to present
data which can be manipulated and explored by the user
- Presentation of results in visual form - integration of the results of machine-
based data mining back into the visualization to show things like coverage,
accuracy, and the interrelationships between rules
- Bridging the human-computer communications gap - producing textual rules
that are simple to understand and showing the effects of those visually, and
allowing the data mining process to iterate
- Exploratory approach - allowing the user to pose questions about differ-
ent parts of the data and follow lines of enquiry, focussing on whatever is
interesting
- Comprehensible rules - tunable to give overviews or detail as required

1.3 Innovation

The innovative parts of the system are the data visualization approach, the
discovery visualization, and the ability to produce rules that can be overviews
(less accurate but painting a broad-brush picture) or specific (precise, and often
more complex) or anywhere in between. However, the key point is that these
approaches are combined into one system whose holistic effect is greater than
the sum of the individual advances.

1.4 Relationship to Ambient Intelligence

One of the characteristics of ambient systems is that they utilize a much larger
proportion of the information and context inherent in data in order to produce
results. The Haiku system works in sympathy with these goals. Since the ap-
proach is iterative, as facts and knowledge about the data are discovered, these
could be feed back into the system to guide further discoveries and results, al-
lowing the system to build upon the knowledge it has created. In addition, the
visualization allows information to be presented to the user, or to a group of
users, allowing them to liaise and collectively explore the space.
Since the system can work in real time, new data could be constantly added
into the system - users are not constrained to working with a fixed data set.
The plasticity of both the visualization approach and the GA-based knowledge
discovery system ensures that if new parameters or features of the current data
set are discovered these can be added in as well. Haiku could therefore utilize
new contextual and environmental information in real time, tapping in to both
machine and human capabilities for adaptively processing changing data.