SCSC2003 Abstract S91675
Towards a Metric Suite for Dicrete Event Trace Validity
Submitting Author: Dr. Levent Yilmaz
Abstract:
Simulation model validation involves substantiating the accuracy of a model with respect to the system or phenomena under study. This requires accurate prediction of the degree of consistency between the model and the system. Statistical validity measures exist for quantitative model validation, but such metrics often examine the input/output and response surface differences to facilitate drawing inferences with regard to model's validity. On the other hand, there is a lack of quantitative metric suites that help determine if the way the model behavior is achieved is as intended and consistent with the actual system processes. This paper proposes a solution technique that facilitates measuring similarity of system and model processes. Sequences of events are important forms of data that characterize such behavioral processes. The proposed approach addresses operational model validity by uncovering and measuring discrepancies among sequences of events observable during system operation and model execution. The metrics are developed using event sequence similarity measures. In particular, optimal sequence transformation operations along with their costs are used to derive the proposed metrics. The optimal sequence of transformation operations are computed using a well-known efficient dynamic programming algorithm.
Back to SCSC2003 Abstracts