Segmentation of Continuous Data Streams Based on a Change Detection Methodology - Advanced Techniques in Knowledge Discovery and Data Mining - page 123

The major contribution of the change-detection methodology as described, is
the introduction of a new methodology for change detection and the implication of
the eight possible changes in the data-mining models. The implication of this
novel method in the field of data stream segmentation is described. These notions
are defined, and a specific change-detection procedure is designed to solve the
change-detection problem, based on a set of statistical hypothesis testing. Also, the
methodology was implemented as a new method of finding segmentation in a data
stream.
As change detection is quite a new application area in the field of classification
models of data mining, many issues are left to be investigated:
1. Implementing voting techniques according to the cause(s) and magnitude(s) of
a change(s) detected in period K for combining several models, such as
exponential smoothing, voting weights based on the CD confidence level,
neglecting old periods or problematic periods, etc.
2. Integrating the change-detection methodology in an existing data-mining
method. As seen, the change-detection procedure's complexity is only O( n ).
One could integrate this procedure with an existing incremental learning
algorithm, which will continue efficiently rebuilding an existing model if the
procedure detects a significant change in the newly obtained data. This option is
also applicable for meta-learning and combining methods.
3. Implementing the methodology in a new search algorithm for finding an
optimal segmentation of a data stream with respect to a variety of data-mining
models.
4. Using the CD statistical hypothesis testing for specific attribute monitoring.

References

[1] Ali, K., and Pazzani, M., Learning multiple descriptions to improve

classification accuracy, International Journal on Artificial Intelligence Tools ,
4, 1995.

[2] Case, J. et al., Incremental concept learning for bounded data mining, Proc.

Automata Induction, Grammatical Inference, and Language Acquisition,
Workshop at the 14th International Conference on Machine Learning (ICML-
97), 1997, and Information and Computation , 152(1), 74-110, 1999.

[3] Chan, P. K., and Stolfo, S. J., A comparative evaluation of voting and meta-

learning on partitioned data, in Proc. 12th Intl. Conf. on Machine Learning ,
90-8, 1995.

[4] Chan, P. K., and Stolfo, S. J., Sharing learned models among remote database

partitions by local meta-learning, in Proc. 2nd Intl. Conf. on Knowledge
Discovery and Data Mining , 2-7, 1996.

[5] Cheung, D., Han, J., Ng, V., and Wong, C. Y., Maintenance of discovered

association rules in large databases: An incremental updating technique, in
Proc. of 1996 Int'l Conf. on Data Engineering (ICDE'96), New Orleans, LA,
1996.