Page 269 -

9.4 Predict 251

Fig. 9.7 Another Declare
specification. Note that c1,
c2, and c3 imply that e
cannot be executed without
permanently violating the
specification

both g and h need to be executed after a,b,d,e . However, the execution of both g
and h results in a permanent violation of c1. Because there is no possible future in
which all constraints are satisfied, the Declare specification is permanently violated
by prefix a,b,d,e and an alert is generated directly after e occurs. Note that in
the latter scenario, there are only temporarily violated constraints whereas the whole
specification is permanently violated. Therefore, advanced reasoning is required to
determine whether an event signifies a deviation or not. As shown in [68, 108], one
can use model checking or abductive logic programming to detect such deviations
and provide informative alerts.

9.4 Predict

The second operational support activity we consider is prediction.Asshown in
Fig. 9.8, we again consider the setting in which users are interacting with some
enterprise information system. The events recorded for cases can be sent to the op-
erational support system in the form of partial traces. Based on such a partial trace
and some predictive model, a prediction is generated. Examples of predictions are:

• The predicted remaining flow time is 14 days.
• The predicted probability of meeting the legal deadline is 0.72.
• The predicted total cost of this case is € 4500.
• The predicted probability that activity a will occur is 0.34.
• The predicted probability that person r will work on this case is 0.57.
• The predicted probability that a case will be rejected is 0.67.
• The predicted total service time is 98 minutes.
In the fictive example shown in Fig. 9.8, the operational support system predicts that
the completion date will be April 25, 2011.
Various techniques can be used to generate predictions. For example, the su-
pervised learning techniques discussed in Sect. 3.1.2 can be used to answer some
of these questions. Using feature extraction, relevant properties of the partial trace
need to be mapped onto predictor variables. Moreover, the feature we would like to