An improved feature ranking method for diagnosis of systematic timing uncertainty

Abstract

For diagnosis of systematic modeling uncertainty, an earlier work proposes a path-based methodology that employs support vector classification analysis to rank so-called delay entities. This work explains that delay entities can be seen as path features that are used to encode the characteristics of a path. We present an improved path feature ranking algorithm based on support vector epsiv-insensitive regression. We also discuss how to check if a dataset is too noisy for the analysis. Experimental results are presented to explain the ranking methodology and demonstrate the effectiveness of the improved approach.