Process Mining (Part 2/3): More on bupaR package
Recap
In the last post, the discipline of event log and process mining were defined. The bupaR package was introduced as a technique to do process mining in R.
Objectives for This Post
- Visualize workflow
- Understand the concept of activity reoccurrences
We will use a pre-loaded dataset sepsis from the bupaR package. This event log is based on real life management of sepsis from the point of admission to discharge.
The dataset has 15214 activity instances and 1050 cases.
library(plyr)
library(tidyverse)
library(bupaR)
sepsis## Event log consisting of:
## 15214 events
## 846 traces
## 1050 cases
## 16 activities
## 15214 activity instances
##
## # A tibble: 15,214 x 34
## case_id activity lifecycle resource timestamp age crp
## <chr> <fct> <fct> <fct> <dttm> <int> <dbl>
## 1 A ER Regi~ complete A 2014-10-22 11:15:41 85 NA
## 2 A Leucocy~ complete B 2014-10-22 11:27:00 NA NA
## 3 A CRP complete B 2014-10-22 11:27:00 NA 210
## 4 A LacticA~ complete B 2014-10-22 11:27:00 NA NA
## 5 A ER Tria~ complete C 2014-10-22 11:33:37 NA NA
## 6 A ER Seps~ complete A 2014-10-22 11:34:00 NA NA
## 7 A IV Liqu~ complete A 2014-10-22 14:03:47 NA NA
## 8 A IV Anti~ complete A 2014-10-22 14:03:47 NA NA
## 9 A Admissi~ complete D 2014-10-22 14:13:19 NA NA
## 10 A CRP complete B 2014-10-24 09:00:00 NA 1090
## # ... with 15,204 more rows, and 27 more variables: diagnose <chr>,
## # diagnosticartastrup <chr>, diagnosticblood <chr>, diagnosticecg <chr>,
## # diagnosticic <chr>, diagnosticlacticacid <chr>,
## # diagnosticliquor <chr>, diagnosticother <chr>, diagnosticsputum <chr>,
## # diagnosticurinaryculture <chr>, diagnosticurinarysediment <chr>,
## # diagnosticxthorax <chr>, disfuncorg <chr>, hypotensie <chr>,
## # hypoxie <chr>, infectionsuspected <chr>, infusion <chr>,
## # lacticacid <chr>, leucocytes <chr>, oligurie <chr>,
## # sirscritheartrate <chr>, sirscritleucos <chr>,
## # sirscrittachypnea <chr>, sirscrittemperature <chr>,
## # sirscriteria2ormore <chr>, activity_instance_id <chr>, .order <int>We will subset a smaller and more granular event log to make illustrations more comprehensible.
activity_frequency(sepsis, level = "activity") %>% arrange(relative)# least common activity ## # A tibble: 16 x 3
## activity absolute relative
## <fct> <int> <dbl>
## 1 Release E 6 0.000394
## 2 Release D 24 0.00158
## 3 Release C 25 0.00164
## 4 Release B 56 0.00368
## 5 Admission IC 117 0.00769
## 6 Return ER 294 0.0193
## 7 Release A 671 0.0441
## 8 IV Liquid 753 0.0495
## 9 IV Antibiotics 823 0.0541
## 10 ER Sepsis Triage 1049 0.0689
## 11 ER Registration 1050 0.0690
## 12 ER Triage 1053 0.0692
## 13 Admission NC 1182 0.0777
## 14 LacticAcid 1466 0.0964
## 15 CRP 3262 0.214
## 16 Leucocytes 3383 0.222sepsis_subset<-filter_activity_presence(sepsis, "Release E") # cases with least common activity to achieve smaller eventlogVisualizing Workflow
In an event log, the sequence of activities are captured either using time stamps or activity instance identifier. The sequential order of activities allows us to create a workflow on how a case was managed. This workflow can be compared against theoretical workflow models to identify deviations. It can also reveal the reoccurrence of activities which will be covered later. The most intuitive approach to examine workflow is with a process map and this is achieved with the process_map function from bupaR
The process map is either created base on frequency of activities (i.e. process_map(type = frequency())) or base on duration of activities (i.e. process_map(type= performance())). I have focused on the frequency aspect which is the default argument.
There are 4 arguments that you can supply to process_map(type = frequency()) which determines the value to be displayed. The values can reflect either the number of activity instances or the number of cases. The values displayed can be in absolute or relative frequency.
Process mapping based on absolute activity instances
sepsis_subset %>% process_map(type = frequency("absolute"))Process mapping based on absolute number of cases for the activity
sepsis_subset %>% process_map(type = frequency("absolute_case"))Reoccurence of Activities
In the above process map, the top of the boxes for “CRP”, “Leucocytes”, “Admission NC” and “Lactic Acid” has an arrow head and its tail belonging to the same activity. These arrows indicate reoccurrence of the activities for the same case. Reoccurrence of activities can suggest inefficiency or interruptions or disruptions which may warrant further investigation to optimise workflow.
bupaR has 2 constructs to define reoccurrence of activities for the same case.
Construct 1 (resource reduplicating the activity)
If the same resource reduplicates the activity for a particular case, it is known as “repeat” in bupaR’s terminology. If a different resource reduplicates the activity, it is known as “redo”.
Construct 2 (activity instances when the activity reoccured)
When the activity for a specific case reoccurs as consecutive activity instances, it is term as “self loop”. This is an example of “1 self loop”
| Activity | Self Loop |
|---|---|
| ER Triage | - |
| CRP | 1 |
| CRP | 1 |
| Release A | - |
When the activity reoccurs as non consecutive activity instances, it is known as “repetition”. In other words, there are other activities that occur before that specific activity is replicated. This is an example of “1 repetition”
| Activity | Repetition |
|---|---|
| ER Triage | - |
| CRP | 1 |
| Leucocytes | - |
| CRP | 1 |
The permutation of these constructs result in these 4 types of activity reoccurrences:
- Redo self-loop
- Redo repetition
- Repeat self-loop
- Repeat repetition
Let’s look at some examples using bupaR
Which activities reoccurred consecutively in a case?
number_of_repetitions(sepsis_subset, level="activity", type="all")## # Description: activity_metric [12 x 3]
## activity absolute relative
## <fct> <dbl> <dbl>
## 1 Admission IC 0 0
## 2 Admission NC 2 0.167
## 3 CRP 6 0.188
## 4 ER Registration 0 0
## 5 ER Sepsis Triage 0 0
## 6 ER Triage 0 0
## 7 IV Antibiotics 0 0
## 8 IV Liquid 0 0
## 9 LacticAcid 3 0.158
## 10 Leucocytes 6 0.136
## 11 Release E 0 0
## 12 Return ER 0 0Which activities reoccurred consecutively in a case where the same resource repeated the activities?
number_of_repetitions(sepsis_subset, level="activity", type="repeat")## # Description: activity_metric [12 x 3]
## activity absolute relative
## <fct> <dbl> <dbl>
## 1 Admission IC 0 0
## 2 Admission NC 0 0
## 3 CRP 6 0.188
## 4 ER Registration 0 0
## 5 ER Sepsis Triage 0 0
## 6 ER Triage 0 0
## 7 IV Antibiotics 0 0
## 8 IV Liquid 0 0
## 9 LacticAcid 3 0.158
## 10 Leucocytes 6 0.136
## 11 Release E 0 0
## 12 Return ER 0 0Which cases had activities duplicated? The duplicated activities were interrupted by other activities and these duplicated activities were redone by a different resource.
number_of_selfloops(sepsis_subset, level="case", type = "redo")## # Description: case_metric [6 x 3]
## case_id absolute relative
## <chr> <dbl> <dbl>
## 1 BCA 0 0
## 2 CY 1 0.0303
## 3 JAA 0 0
## 4 JM 1 0.0769
## 5 LG 1 0.0244
## 6 SAA 0 0Summing Up
In this post, we learnt how to visualize the workflow registered in event logs with a process map. We also learnt bupaR’s constructs of activity reoccurrences and therefore the types of activity reoccurrences. If you want to learn more about bupaR, head over to Datacamp for more comprehensive and interactive lessons. P.S. I’m not sponsored nor affiliated to Datacamp
In the next post, we’ll look at more visualizations for process analysis.