Transcript Natalia Sidorova The classical setting for process mining

Mine your life!
Natalia Sidorova
The classical setting for process mining:
monitoring business processes
Event
Event
Log
Event
Log
Log
Operational support
(predictions,
recommendations,
…)
Mining
Compliance
checking
Analysis
The classical setting for process mining:
Monitoring business processes
We can apply process mining to
• understand the process,
• check compliance,
• generate predictions,
• …
We can tell at every
moment:
• where each
package is in the
process,
• what is in each
trolley,
• who is doing
what
Mining in a smart home environment?
Mining in a smart home environment?
May 27, 2015,18:05:35: start bottle movements
May 27, 2015, 18:05:45 start increasing the bottle weight
May 27, 2015,19:05:35: start bedtime May 27, 2015, 18:06:02 complete increasing the bottle
May 27, 2015, 19:30:02 light sleep
weight, 200 g
May 27, 2015, 20:15:28 deep sleep
May 27, 2015, 18:07:02 start warming up, 20C
May 27, 2015, 21:15:28 light sleep
May 27, 2015, 18:08:15 stop warming up, 37C
….
…
Physiological signals and activity monitoring
•
•
•
•
Heart rate
Temperature
Light
…
A different setting – same problems:
compliance checking, alignment
• Given a process model for a
“normal baby sleep”, identify
deviations from the model in
the log
• Given definitions of typical
“undesired behavior”, i.e. feeding
the baby with cold milk, signal it at
run-time as early as possible
• Note: the log contains events of
the bottle, not of the baby!
Identifying events in sensor data:
pure data mining is not always enough
or
Time is important
• Absolute time and its abstractions:
• Between 9 pm and 6 am
• On Sundays
• …
• Relative time
• Time from the last feeding
• Time from falling asleep
• But also combinations of the two:
• Time from the last feeding till 6am
Small big data
What is a case?
•
•
•
•
A day? A week? A year?
A person?
A household?
…
Mulitiple lifecycles interacting with each
other
Getting right data
• Drivers data on road and in the simulator: GPS log,
accelerometer, heart rate, activities, sleep, …
• Meaningful events?
• Baby sleep logs
• Other events?
• Smart bottle data
• Other events?
• Coffee corner data
• Difficult to identify events belonging to the “same case”
• Data from sensors at home
• Still to be obtained…
On the positive side
• Lots of interest from industry
• Fast developments in the Quantified Self area
• New challenges in process mining
• Time as a first class citizen
• Hierarchical mining
• Mining with “ambiguous” labels