||U.S. Air Force (Rome Labs)
||Much of the research in the area of Knowledge Discovery in Databases (KDD) has focused on the development of more efficient and effective data mining algorithms, but issues related to the usability of these techniques in extracting exploitable knowledge from databases has drawn significant attention. Since the exploratory process associated with KDD proceeds in a data-driven manner, it is crucial that these tools are seamlessly integrated so as to allow flexible utilization of tools and operation chaining. Especially useful to intensive knowledge discovery processes is the ability to intuitively utilize the results of data mining operations in subsequent exploration steps. To date, such capabilities have largely been neglected to the significant detriment to users.
- IKODA (Intelligent KnOwledge Discovery Assistant) utilizes “data aware” visualization techniques that lie atop a unified and persistent knowledge representation and provide a mapping between graphical objects and the underlying data resources. This approach results in the ability to perform direct manipulation operations such as drag-and-drop transfer of data between tools and a unique capability to explore data mining results. Unlike previous integrated KDD systems, IKODA’s visualizations act as interactive tools rather than simple information displays. Specifically, IKODA’s visualizations of data mining results (e.g., decision trees and automatically created data clusters) can be manipulated and used directly to form new datasets that feed future data mining operations. The resulting “recursive” knowledge discovery capability represents a substantial step forward in reducing KDD tool complexity while simultaneously increasing flexibility and efficacy.
- The Predictive Model Markup Language (PMML) is an XML-based language which provides a quick and easy way to define and share predictive models between applications. PMML’s open text-based format enables researchers and commercial users to carry out different data mining tasks (e.g. train, test, apply, visualize) with different tools and, if necessary, edit the model (as an XML document) using a simple text editor. IKODA supports a PMML interface that serves as an API to IKODA’s data mining. PMML
- A general data mining solution is not always the best solution to a domain-specific problem. Many users would like to utilize the functionality of data mining algorithms but are not database experts or statisticians. They want an interface that uses the domain terminology and hides the complexity of the data mining algorithms. The PMML interface makes it possible to use IKODA as a data mining engine to power vertical solutions. The data mining algorithms work behind the scenes of the customized front-end and the user receives the results in the terms to which they are accustomed.
||The system has been in use in Stottler Henke”s consulting work since 2000, both as a stand-alone application and as a back end data mining engine for specialized applications.
||The functionality provided by IKODA has provided a solid foundation for related applications to build upon. We have used it as a major component in applications as varied as database fraud detection, network anomaly detection, and epidemiological research.