Dealing with uncertainty in the real-time knowledge discovery process

Abstract

This paper will examine where uncertainty may lie in the knowledge discovery process through the use of case studies in disaster management, in turn leading to a discussion of what future action is required to address the uncertainty that may lie within knowledge obtained through these techniques. We describe our approach to address three types of issues: accuracy, efficiency, and usability. Typically, data mining techniques have higher false positive rates than traditional data exploratory methods, making them unusable in real-time systems. Also, these techniques tend to be inefficient (that is, computationally expensive) during the steps of a knowledge discovery process, particularly during training and evaluation. This prevents them from being able to process data and detect anomalies, hot-spots, or patterns in real-time applications. Finally, disaster management applications require large amount of training data and are significantly more complex than traditional GIS applications. These problems are inherent in developing and deploying any real-time data mining based system, and although there are trade-offs between these three groups of issues, each can generally be handled separately. The paper concludes by presenting the key design elements for supporting a real-time knowledge discovery process and group them into which general issues they address. © 2005 Springer-Verlag Berlin Heidelberg.