Using decision tree learning to predict workflow activity time consumption

Activity time consumption knowledge is essential to successful scheduling in workflow applications. However, the uncertainty of activity execution duration in workflow applications makes it a non-trivial task for schedulers to appropriately organize the ongoing processes. In this paper, we present a K-level prediction approach intended to help workflow schedulers to anticipate activities' time consumption. This approach first defines K levels as a global measure of time. Then, it applies a decision tree learning algorithm to the workflow event log to learn various kinds of activities' execution characteristics. When a new process is initiated, the classifier produced by the decision tree learning technique takes prior activities' execution information as input and suggests a level as the prediction of posterior activity's time consumption. In the experiment on three vehicle manufacturing enterprises, 896 activities were investigated, and we separately achieved and average prediction accuracy of 80.27%, 70.93% and 61.14% with K = 10. We also applied our approach on greater values of K, however the result is less positive. We describe our approach and report on the result of our experiment.