State Transition Diagram Dependency Detection

OF THESIS STATE TRANSITION DIAGRAM DEPENDENCY DETECTION I present an algorithm that builds state transition diagrams out of event traces, in order to find causal relationships between the various events in these traces. The main application of this algorithm is high-level debugging (for situations where it is difficult or impossible to replicate a specific instance of a failure). But many other uses, such as market prediction, credit card fraud tracking, and data mining, are also possible. The algorithm is the latest in a family of statistics-based techniques for modeling process behavior, called Dependency Detection. It collects relatively short, significant sequences (snapshots), to generate an integrated, abstract overview model of the analyzed process. Also, detailed performance and accuracy evaluations of the algorithm are presented. Gabriel L. Somlo Department of Computer Science Colorado State University Fort Collins, Colorado 80523 Fall 97