Activity duration prediction of workflows by using a data science approach: Unveiling the advantage of semantics

Abstract

Organizations often have to face a dynamic market environment. Processes must be frequently adapted in order to stay competitive and allow an efficient workflow. Data Science approaches are currently often used in analysis methods to identify influential indicators on processes and learn predictive models to estimate the duration of an activity. However, current methods do not or only partially make use of semantic information in process analysis. The results are unprecise or incomplete, because not all influential indicators have been unveiled and therefore used in the predictive models. We want to make use of the semantics and show the advantage by applying them on existing data science methods for predicting the duration of an activity in a process. Therefore, we 1) enrich process data with meta-information and background knowledge 2) extend existing data science methods so that they include semantic information in their analysis and 3) apply data science methods for predicting values and compare the results with methods, which do not use semantics.