Decision rules play a crucial role in business process execution. Knowing and understanding decision rules is of utmost importance for business process analysis and optimization. So far, decision discovery has been merely based on data elements that are measured at a single point in time. However, as cases from different application areas show, process behavior and process outcomes might be heavily influenced by additional data such as sensor streams, that consist of time series data. This holds also true for decision rules based on time series data such as 'if temperature > 25 for more than 3 times, discard goods'. Hence, this paper analyzes how time series data can be automatically exploited for decision mining, i.e., for discovering decision rules based on time series data. The paper identifies global features as well as patterns and intervals in time series as relevant for decision mining. In addition to global features, the paper proposes two algorithms for discovering interval-based and pattern-based features. The approach is implemented and evaluated based on an artificial data set as well as on a real-world data set from manufacturing. The results are promising: the approach discovers decision rules with time series features with high accuracy and precision.

Top