It looks like you're new here. If you want to get involved, click one of these buttons!
Announcements / News
- BPI Challenge 2020
- BPI Challenge 2019
- BPI Challenge 2018
- BPI Challenge 2017
- BPI Challenge 2016
In this Discussion
Remaining time prediction
what is the best method in prom for remaining time prediction?
Thanks in advance
remaining time prediction
edited June 2020
Assuming that you start from an event log, you could first mine a transition system using the "Mine Transition System" plugin, and second analyze the event log and the mined transition system using the "Analyze Transition System" plugin. This extends the transition system with timing information, among which the remaining times. If you select a state in the transition system, the remaining time will be shown in the table below.
If you want to predict the remaining time for a trace which is still on-going, you can replay the trace in the transition system. The state you end up in, gives you a prediction for the remaining time.
In the first (mining) step, you can configure which information is important to distinguish the different states. By default, only the last action leading up to a state will determine the state, but you could extend this in different ways:
- You could also look at actions that follow a particular state ("forward" instead of "backward").
- You could also look too more multiple actions instead of just one.
- If you consider multiple actions, you could also configure the way they are collected:
-- as a set (order and cardinality do not matter),
-- as a multi-set (order does not matter, but cardinality does), or
-- as a list (both order and cardinality matter).
- You could also include the value of attributes as given in the log.
This defines the set of states you are interested in, and the analysis will take this into account.
There could be other ways to do this, so I cannot claim this is the best way, but it is a way.
Post edited by hverbeek on
Forum Software Powered by Vanilla