Next: Acknowledgments Up: Model-Based Mining of Previous: Model Revision

Summary

As high resolution remotely sensed data such as satellite images become increasingly available, a wide range of new opportunities for using this data to monitor and predict environmental factors such as air pollution and epidemic diseases begin to emerge. These remotely sensed data are usually taken on a periodic basis, and provide a comprehensive coverage of the ground. Consequently, these data enables near real-time monitoring of many environmental factors simultaneously.

In this chapter, we have investigated a model-based data mining framework for extracting, validating, and refining environmental diseases based on the availability of these remotely sensed data. In contrast to more conventional approaches, which rely on either passive or active surveillance, the proposed framework

decomposes an environmental event such as the spreading of Hantavirus Pulmonary Syndrome (HPS), Lyme disease, and Dengue fever, using an object-oriented approach,
validates the model by efficiently evaluating the candidates in the database, and
revises the model through iterative refinement based on nonlinear multidimensional scaling.

The main advantages of the proposed framework includes

Integrating the development and the deployment of environmental models on the same platform;
Allowing the user to develop models in an intuitive fashion by using a drag-and-drop model development user interface;
Evaluating both localized and global models efficiently by taking advantage of the efficient query support for composite objects;
Allowing the user to revise the model interactively by adopting a relevance feedback algorithm based on nonlinear multidimensional scaling.

This framework is currently being implemented and we expect to report the results in a future paper.