EFFICIENT METHOD FOR DETECTING AND LOCALIZING CONCEPT DRIFTS FROM EVENT LOGS IN PROCESS MINING

Abstract: A generic framework and speci?c techniques to discover once a process changes and to localize the elements of the process that have modified. Totally different options are projected to characterize relationships among activities. These options are wont to discover variations between serial populations. The drift might even be periodic (e.g., due to seasonal in?uences) or one-of-a-kind (e.g., the results of latest legislation). Projecting the information onto a random lower-dimensional mathematical space yields results love standard spatial property reduction ways like principal part analysis: the similarity of knowledge vectors is preserved well below random projection. Random projections (RP) is computationally signi?cantly more cost-effective than mistreatment, e.g., principal part analysis. RP employing a distributed random matrix provides extra machine savings in random projection.

Keywords: Concept drift, ?exibility, hypothesis tests, random projection, dimensionality reduction, image data, text document data, high-dimensional data.