A Knowledge Discovery Framework for Civil Infrastructure Contexts

Carnegie Mellon University, Pittsburgh PA

Investigators

Abstract

Proposal: CMS-9987871 PI: James Garrett Institution: Carnegie Mellon University Date: April 20, 2001 ABSTRACT CMS-9987871 " A Knowledge Discovery Framework for Civil Infrastructure Contexts" PI: Garrett, James, Carnegie Mellon University; Christos Faloutsos, Carnegie Mellon University; and Sue McNeil, University of Illinois at Chicago. The primary objective of this research is to specialize the abstract CRISP-DM (Cross-Industry Standard Process for Data Mining) process model being developed within the Knowledge Discovery in Databases (KDD) community, into a framework for use in civil infrastructure problem domains. With the recent accumulation of domain data, civil infrastructrue researchers have turned to data-intensive techniques to aid their understanding of deterioration mechanisms and usage patterns. During roughly the same time period, researchers in the machine learning, database, and statistical communities began to develop a set of new tools and techniques, known as CRISP-DM, to analyze very large databases. This framework will assist civil infrastructure researchers in systematically applying the CRISP-DM process for their data analysis needs. Such a framework will become vital for civil infrastructure researchers as they begin to analyze the enormous amounts of infrastructure data they have collected. The research team will identify the analyze preliminary case studies in civil infrastructure using the CRISP-DM process. Case studies will be chosen based on the uniqueness of their KDD problem characteristics. Special attention will be paid to those case studies that present challenging issues in data quality and data preparation, the most time-consuming and difficult stages of the process as well as the least-studied. The research team will classify civil infrastructure data analysis needs in terms of CRISP-DM process characteristics. All phases of the process will be addressed, but the data understanding (including data quality), data preparation, and modeling phases will be treated in-depth. The research team will then develop a more specific framework for applying the CRISP-DM process to civil infrastructure analysis needs. The research team will also identify and conduct case studies with which to validate the framework. Finally, the development and deployment of a web-based course and a web repository based on this research will be completed during the final year.