Time Prediction Based On Process Mining Taking Concept Drift Into Consideration

The goal of process mining is to extract process-related information by using the event data of some enterprise system. It allows for the automated discovery of process models from event logs. These models provide insights and enable various types of model-based analysis. However, as process mining techniques are becoming more and more mature and the complexity of business models themselves, people are no longer satisfied with static testing or evaluation and more desired to use the process models at running time. For example, one may want to know the probability of execute one activity in the next step, the best candidate for the implementation of the activity, or if the case can complete on time and the remaining time if it can do it. We see this referred to as the prediction service. As well as this paper proposes a new method that we can do time prediction of the running process by extending the discovered model with information. In real life we often need practical and reliable time prediction information. For example, when a customer asks for the arrival time of her online shopping goods, the seller should give a relatively accurate answer.Now many scholars have made a great contribution on predicting the completion time of running instances and a lot of algorithms have been proposed, but they mostly ignored concept drift which means the influence of the external factors. In order to improve the accuracy of the prediction, we take the concept drift into account on the basis of the previous algorithm. Obviously, it is quite important to make relatively accurate predictions in real life. With regards to this, we can predefine some social or external factors that may influence the handing of a case, then do cluster analysis on the annotated transition system. We can make a division of the current execution state through the use of K-means algorithm, and then determine which category it belongs to dynamically. According to the integrated prediction function we defined in this paper, we can do time prediction in a more meaningful way.Our approach has the theoretical innovation through combining concept drift and time prediction. And experiments show that our algorithm performs better than simple heuristics and regressions models whether using the abstract cases or the real event log of process as a input. It also has a considerable degree of improvement over the previous when considering concept drift.