Skip to content

Latest commit

 

History

History
935 lines (586 loc) · 60 KB

README.md

File metadata and controls

935 lines (586 loc) · 60 KB

Carried out by:

Giuseppe Aniello

Domenico Armillotta

Leonardo Bellizzi

Eodardo Malaspina

1. LANDSCAPE

(All)

2. BPMN Diagrams

2.1. CONFIGURE VEHICLE DAMAGE DETECTION (ALL)

2.2. PREPARE VEHICLE IMAGES (Domenico Armillotta)

REPORT is composed by:

Images(convolutional feature,

Gestional Data (plate, category, driver age)

Black Box data (gps,acceleration)

Report Components make up Report

First Missing value check is done by machine, if there are other missing values, they will be checked by a human.

2.3. GENERATE PREPARED REPORT SET (Leonardo Bellizzi)

2.4. DEVELOP VEHICLE DAMAGE DETECTION (Giuseppe Aniello)

2.5. DETETECT VEHICLE DAMAGE (Edoardo Malaspina)

2.6. CHECK VEICHLE DAMAGE DETECTOR (Edoardo Malaspina)

3.SALARIES

ACTOR LINK COST($/H) NORMALIZED SALARY
Clerk https://www.indeed.com/career/clerk/salaries 15,8 1
Technician https://www.indeed.com/career/technician/salaries?from=top_sb 20,5 1,2
Damage AssessorIn Transportationindustry https://www.glassdoor.com/salaries/damage-assessor-salary 42,5 2,6
Data analyst https://www.prospects.ac.uk/job-profiles/machine-learning-engineer 32,1 2,0
Machine LearningEngineer https://www.indeed.com/career/machine-learning-engineer/salaries 55.74 3.5

4. USE CASES

4.1. CRM (Domenico Armillotta)

CLERK:CUSTOMERS REGISTRATION

SUBTASK ACTOR ACTION COGNITIVEEFFORT OCCURRENCE COST
1 CLERK CLERK open customer registrationinterface 1 1 $1 ^ { * } 1 ^ { * } 1$
2 System System shows new customerregistration interface
3 CLERK Clerk choose 'add new customer' 1 1 $1 ^ { * } 1 ^ { * } 1$
4 System System shows new customerregistration module
5 CLERK Clerk insert new costumer data (figure1) 1 (Remember) 1 $1 ^ { * } 1 ^ { * } 1$
6 System System adds new customer to DB
Total Cost Total Cost Total Cost Total Cost Total Cost 3
Customers registration
damanicoamifuna@gnal.com [hon condhviso)Carbia a00ountCampo abbligatoo
Nome CogomeLa tua rporta
OttaLa tue risposta
Corgary I0LatMahne Pat SLage000$\textcircled \cdots \cdots$ug
cO ntyOCar einCtyCH Boadpen Dgt

Figure 1:Customer registration form

Users' data that must be inserted into the system by clerk, are taken by a pre compiled form.

SUBTASK ACTOR ACTION COGNITIVEEFFORT OCCURRENCE COST
1 CLERK CLERK opens list of expert interface 1 1 111
2 System System shows list of experts
3 CLERK Clerk chooses experts 1 1 $1 ^ { * } 1 ^ { * } 1$
4 System System update the expert status in 'notavailable'
Total Cost Total Cost Total Cost Total Cost Total Cost 2

4.2. ANNOTATION SYSTEM (Domenico Armillotta)

DAMAGE ASSESSOR IN TRANSPORTATION INDUSTRY: EVALUATE AND CREATE LABELS

SUBTASK ACTOR ACTION COGNITIVE EFFORT OCCURRENCE COST
1 DamageAssessor DamageAssessor openon interface ofthe car underexamination 1 1 112,6
2 System System showsthe result ofthe photo:4photo,one foreach side ofcar(figure1)
3-FOR For each image
3.1 DamageAssessor DamageAssessorEvaluate thedamage andassign a score(1-3) 3 (Analyze) rules:Press 1 if there is no imperfectionPress 2 if there is at least one scratch.(figure 2)Press 3 if there are more than onescratch of damaged part. (figure 3) 4 $3 * 4 * 2 , 6$
4 System The final score isthesum of 4score
Total Cost Total Cost 33,8

0 2 Assign Label 3

Figure 2:Labels assignment

Figure 4 : Example of scratch on a car door

Figure 5 : Example of several damages an a rear car door

4.3. INGESTION SYSTEM (Domenico Armillotta)

DATA ANALYST: CONFIGURE INGESTION SYSTEM

SUBTASK ACTOR ACTION COGNITIVEEFFORT OCCURRENCE COST
1 DataAnalyst Data Analyst open the interface to setparameters 1 1 211
2 System System shows the interfaces
3 DataAnalyst Data Analyst set the threshold of missingvalue for the system 1 (Remember) 1 $1 ^ { * } 2 ^ { * } 1$
4 DataAnalyst Data Analyst set the condition of"complete" REPORT 1 (Remember) 1 121
Total Cost Total Cost Total Cost Total Cost Total Cost 6

DATA ANALYST: START INGESTION SYSTEM

SUBTASK ACTOR ACTION COGNITIVEEFFORT OCCURRENCE COST
1 System System is ready to be started
2 DataAnalys Data Analyst presses “start system”button 1 1 $2 ^ { * } 1 ^ { * } 1$
Total Cost Total Cost 2

4.4. PREPARATION SYSTEM (Domenico Armillotta)

DATA ANALYST: CONFIGURE PREPARATION SYSTEM

SUBTASK ACTOR ACTION COGNITIVEEFFORT OCCURRENCE COST
1 DataAnalyst Data Analyst open the interface to setparameters 1 1 211
2 System System shows the interfaces
3 DataAnalyst Data Analyst set the preferences for thesystem to correct missing samples 1(Remember) 1 $1 ^ { * } 2 ^ { * } 1$
4 DataAnalyst Data Analyst set the threshold to detectthe absolute outliers 1(Remember) 1 $1 ^ { * } 2 ^ { * } 1$
5 DataAnalyst Data Analyst set the parameters toperform the feature extraction 1(Remember) 1 $1 ^ { * } 2 ^ { * } 1$
Total Cost Total Cost Total Cost Total Cost Total Cost 8

DATA ANALYST: START PREPARATION SYSTEM

SUBTASK ACTOR ACTION COGNITIVEEFFORT OCCURRENCE Cost
1 System System is ready to be started
2 DataAnalyst Data Analyst presses “start system"button 1 1 $2 ^ { * } 1 ^ { * } 1$
Total Cost Total Cost 2

ML ENGINEER: EVALUATE NUMBER GENERATION OF GRADIENT PLOT

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 System shows the plot System
2 ML engineer observesthe flatness of the plot.(Figure 1) MLEngineer Apply(3) 1 13.53
3-IF Plot suffers of flatness. System
3.1 ML Engineer presses thebutton to reduce by onethird the number ofiterations. MLEngineer Remember(1) 1 13.51
4-ELSE Plot does not suffers offlatness. System
4.1 ML Engineer presses thebutton to enlarge by onethird the number ofiterations. MLEngineer Remember(1) 1 $1 ^ { * } 3 . 5 ^ { * } 1$
5 Number of iterations iscorrectly set System
TOTAL= 17.5

Figure 6:Descent gradient plot

ML ENGINEER: EVALUATE TESTING REPORT

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 ML Engineer presses thebutton to generate theevaluation testing report. MLEngineer Remember(1) 1 $1 * 3 . 5 * 1$
2 System generates a reportthat contains the top 5Neural Network withhyperparameters. System
3 ML engineer computes thedifference between errortraining and validationerror and compare it witha threshold. MLEngineer Apply(3) 1 $1 * 3 . 5 * 3$
4-IF The difference is greaterthan threshold System
4.1 ML Engineer presses thebutton to discard theclassifier. MLEngineer Remember(1) 1 $1 * 3 . 5 * 1$
5-IF The difference is smallerthan threshold System
5.1 ML Engineer presses thebutton to maintain theclassifier. MLEngineer Remember(1) 1 $1 * 3 . 5 * 1$
6-ELSE System obtains the validclassifiers
TOTAL= 21
Neural Network Error Training(%) Error Validtion(%)
NN_1 4,3 7.7
NN_2 5.7 4,7
NN_3 1,2 3,4
NN_4 0.7 1.1
NN_5 8,9 7.2

Figure 7. Top 5 classifiers of the report.

ML ENGINEER: CONFIGURE SYSTEM

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 System receives request ofconfiguration System
2 ML Engineer sets number ofneurons MLEngineer Remember(1) 1 $1 * 3 . 5 * 1$
3 ML Engineer sets number oflevels MLEngineer Remember(1) 1 13.51
4 ML Engineer sets threshold MLEngineer Remember(1) 1 $1 ^ { * } 3 . 5 ^ { * } 1$
5 System receives newhyperparameters 1
$T o t a \vert =$ 10.5

ML ENGINEER: START SYSTEM

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 System is ready to be started System
2 ML Engineer presses "startsystem" button MLEngineer Remember(1) 1 $1 * 3 . 5 * 1$
Total = 3.5

4.6. SEGREGATION SYSTEM (Leonardo Bellizzi)

DATA ANALYST: CHECK DATA QUALITY ON RADAR DIAGRAM

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 Data analyst open theRadar Diagram DataAnalyst Remember(1) 1 $1 ^ { * } 2 ^ { * } 1 = 2$
2 System shows the RadarDiagram System $1 * 2 * 2 = 4$
3 Data analyst checks qualityof the radar diagram Dataanalyst Understand(2) 1
4 Data analyst checks if thereare some lacking data Dataanalyst Apply (3) 1 $1 * 2 * 3 = 6$
5-IF(20%) IF there is a lack of data System
5.1 Data analyst press"Reconfiguration" button Dataanalyst Remember(1) 1 $1 ^ { * } 2 ^ { * } 1 = 2$
6-ELSE(80%)
6.1 Data analyst press “Datasetpartition" button Dataanalyst Remember(1) 1 $1 ^ { * } 2 ^ { * } 1 = 2$
TOTAL 16

Radar diagram

Figure 8:Radar diagram

Radar diagram for segregation system (features): Crash speed, Street category (numerical category), Crash G acceleration, Vehicle impact area (numerical category)

DATA ANALYST: CONFIGURE SYSTEM

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 System receives request ofreconfiguration System
2 Data analyst receives request ofa new configuration DataAnalyst 1 1 $1 ^ { * } 2 ^ { * } 1$
3 Data analyst sets qualitythreshold for the Radar Diagram DataAnalyst 4 1 $1 * 2 * 4$
4 System receives newhyperparameters System
$T o t a \vert =$ 10

DATA ANALYST: START SYSTEM

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 System is ready to be started System
2 Data Analyst presses "start system"button DataAnalyst 1 $1 ^ { * } 2 ^ { * } 1$
Total = 2

DATA ANALYST: ANALYZE CLASSIFICATION REPORT

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 System shows monitoring systemreport System 1
5 Data analyst read monitoringsystem report DataAnalyst Understand(2) 1 $2 ^ { * } 2 ^ { * } 1$$= 4$
6-IF IF the two thresholds are therespected System 1
6.1 Data Analyst presses "configuredevelopment mode" button DataAnalyst Remember(1) 1 $2 ^ { * } 1 ^ { * } 1$$= 2$
7 - ELSE --- System
7.1 Data Analyst presses "classifier isworking correctly" button DataAnalyst Remember(1) 1 $2 ^ { * } 1 ^ { * } 1$$= 2$
Total = 8

DATA ANALYST: CONFIGURE SYSTEM

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 System receives request ofconfiguration System 1
2 System open the monitoringinterface System 1
3 Data Analyst set the threshold forconsecutive errors DataAnalyst Remember(1) 1 $1 ^ { * } 1 ^ { * } 2$$= 2$
4 Data Analyst set the threshold fortotal errors DataAnalyst Remember(1) 1 $1 ^ { * } 1 ^ { * } 2$$= 2$
$T o t a \vert =$ 4

DATA ANALYST: START SYSTEM

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 System is ready to be started System 1
2 Data Analyst presses “startsystem" button DataAnalyst Remember(1) 1 $1 ^ { * } 2 ^ { * } 1$$= 2$
$T o t a \vert =$ 2
Monitoring System report Monitoring System report Monitoring System report
Expert labelclass Damagedetector class Result
1 2
2 2
3 3
3 1
2 2

5. AS IS - TO BE MODEL

5.1. AS IS MODEL (AII)

5.1.1. ASSUMPTIONS TO BUILD UP GATEWAYS PROBABILITIES:

5 classifiers are trained and each one is retrained 3 timed. For each training 200 tokens are requested.

For training phase 3000 tokens have been counted on 10000 total tokens (30% for training -66% for execution and 4% for monitoring).

For the window of monitoring a window of 20 classifications every 500 executions has been follows.

A new customer is inserted every 50.

10000 reports are acquired, and 3000 of them are used for training so performing under sampling a probability of have balanced data is 80/20% (in the TO BE will be 100/0).

Small companies may not have enough data report so the probability is 90/10 (in the TO BE will be 100/0).

Considering that images are taken by experts, black boxes are quite reliable, and the task of the clerk is simple so 3% of missing values above the threshold are assumed.

It's assumed that after the training of the first classifier the number of generations is faster to find so in general so 95% of "number of generations properly set" is assumed.1% probability of having a wrong radar diagram, is assumed, for each feature so we assume a 4% overall.

In the grid search 2 hyper parameters are set so different combinations of 5 possible values for the first hyperparameter are assumed and 10 for the second so 50 tries for 5 classifiers for 3 training each. Overall 7.5% probability must be in the grid search.

5.1.2. Gateway probabilities

GATEWAYS PROBABILITIES (%) PROBABILITIES (%)
YES NO
IS NEW CUSTOMER? 2 98
ARE DATA REPORT BALANCED? 80 20
ARE MISSING COMPONENTS ABOVETHRESHOLD? 3 97
ARE NUMBER OF GENERATION PROPERLY SET 95 5
ARE WE IN THE WINDOW OF MONITORING? 4 96
IS IN THE EXECUTION PHASE? 70 30
IS RADAR DIAGRAM, CORRECT? 96 4
IS THE REPORT COMPLETED? 90 10
IS THE REPORT COMPLETED? 90 10
IS THE VALIDATION VIA GRID CYCLE? 7.5 92.5
IS THE CLASSIFIER WORKING CORRECTLY? 95 5
IS THE CYCLE FINISHED? 60 40
IS THERE ENOUGH DATA REPORT? 90 10

5.1.3. Statistic AS IS

BIMP

BMP is a fat and simple wob-baned uner interface to simulate business process models uing the QBP Simulator.

See the grting stated puide to mad more about the featuns,BP can te uoed for toe for acadeik and tal purpooes.Choose the venion boe

Acadiemic Ta Mambers

BIMP-Academic

Academic version of BMP is suppored by Uriversity of Tartu and the Esterian Reseerch Counc

Ai PN

BPMN SMULATION A8 18.hpmm-

BPN Dagram wth resus heat mp Sove ress Dearioad CSV Seves Back t e das

Simulation Results

General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000 General informationComplitad procm ineances 10000
Total co D EUR Total co D EUR Total co D EUR Total co D EUR Total co D EUR Total co D EUR Total co D EUR Total co D EUR Total co D EUR Total co D EUR Total co D EUR
Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge Total simation time 706 weeksChartsProsa eyi gof om Pric c goo Pric sin on12 78P0 1D Braa10)Scenario StatisticsNamum Mamuam Awoge
Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s Process instance cycle times induding off-timesatie hours 13 minutes 142minutes 22minuesProces inceescaing offtimetahou 13 minutes 142mnutes 22mm/s
Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR Process instance casts 0ER OBR OEUR
Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold Activity Durations,Costs,Waiting times,Deviations from TresholdsName Waiting time Duration Duration over threshold Cest Cosl ower threshold
Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom Count Mh A Mo Mim AU Mom Km Pmg Max Mn M Mom Nio Mom
CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0% CLEROCSAOCETRTION OF EPETS CHED 10000 0% 0% 15% 22% 0%
CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0% CERKS ADCUSTONER RGSRATION 163 0% 27 33% 0%
DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372 DAMAGE ASSESSORG ADDMAGE ASSESSMINT 13500 0% 304% 333% 372
DATA ANAOSTSAACLASSCATION REPORT 206 6% 0%DATA ANACYSTS-GDOHEOC REPORT QUAUITY WITH RADAR DAGRANGMD 2522 0% 0% DATA ANAOSTSAACLASSCATION REPORT 206 6% 0%DATA ANACYSTS-GDOHEOC REPORT QUAUITY WITH RADAR DAGRANGMD 2522 0% 0% 12% Es 12% Es 12% Es 0% 0% 0% 0% 0% 0%
DATA ANAOSTSAACLASSCATION REPORT 206 6% 0%DATA ANACYSTS-GDOHEOC REPORT QUAUITY WITH RADAR DAGRANGMD 2522 0% 0% DATA ANAOSTSAACLASSCATION REPORT 206 6% 0%DATA ANACYSTS-GDOHEOC REPORT QUAUITY WITH RADAR DAGRANGMD 2522 0% 0% 1445 36% 1765 36% 1765 0% 0% 0% 0% 0% 0%
DATAANADSTSASOCONFGURE SYSTEM 13500 61 0% DATAANADSTSASOCONFGURE SYSTEM 13500 61 0% 9% 10% 11s 05 05 05 05 05 05 05 05 05 05 05 05 05 05 05 05 05 05
DTLANADST& ASDCONFIGURE SYSTEM6-4O 13500 05 DTLANADST& ASDCONFIGURE SYSTEM6-4O 13500 05 36% 4.45
DATA ANAOYST6 ASOCONFIGURE SYSTEM 13500 05 05 DATA ANAOYST6 ASOCONFIGURE SYSTEM 13500 05 05 725 885
DATA,ANACISTS4DDCONFIGURE SYSTEM 13500 es 05 DATA,ANACISTS4DDCONFIGURE SYSTEM 13500 es 05 54s 665
DATAANACIST&ADOSTART SISTEM 13500 05 DATAANACIST&ADOSTART SISTEM 13500 05 18% 22% 05 05 05 05 05
DATAANALYST&4DSTANT SISTEM 13500 Os DATAANALYST&4DSTANT SISTEM 13500 Os 18s 2s 22s 0s
DATAANACIST&DOSTANT SYSTEM 13500 05 DATAANACIST&DOSTANT SYSTEM 13500 05 18% 22%
DATA ANACIST&40O.STAN SFSTEMG-ALD 13500 0% 0s DATA ANACIST&40O.STAN SFSTEMG-ALD 13500 0% 0s 18% 2% 22% DS 0% 0% 0% 0%
MLENGNEERGIDCONUESSTEN 13500 0% MLENGNEERGIDCONUESSTEN 13500 0% 95% 105% 115%
ML INGIEERS AIDEVRLIATE AND SET NUMER OF GENESIATION OF GRADIENT PLOT 0% 253% 175% 133%
MLENGINEERG FIDEVRLIATE TESTING REPORT 2402 0% 119% 21 23.1% 0%
MLINGNIRG RIDSTART SFSTEM 13500 O% 12% 35% 39% Ds 0% 0

5.1.4.1. Duration

5.1.4.2. Counts

5.2. TO BE MODEL (AII)

5.2.1. Gateway probabilities

GATEWAY PROBABILITIES (%) PROBABILITIES (%)
YES NO
IS NEW CUSTOMER? 2 98
ARE MISSING COMPONENTS ABOVETHRESHOLD? 3 97
ARE NUMBER OF GENERATION PROPERLYSET 95 5
ARE WE IN THE WINDOW OF MONITORING? 4 96
IS IN THE EXECUTION PHASE? 80 20
IS RADAR DIAGRAM, CORRECT? 96 4
IS THE REPORT COMPLETED? 90 10
IS THE REPORT COMPLETED? 90 10
IS THE VALIDATION VIA GRID CYCLE? 7.5 92.5
IS THE CLASSIFIER WORKING CORRECTLY? 95 5
IS THE CYCLE FINISHED? 60 40

5.2.2. Statistics TO BE

BIMP

BMP is a fast and simple web-based user interface to simulate business process moddls using the QBP Simulator.

Soe the gitting staced suithe to read more aboud the features, tee can be uaad for hoe for academis and thad porposes.Chense the version bdem

Acadenic 1납 Merbes

BIMP-Academic

Acadensc vesion of BiP I supponed by Univensiy of Tatu and the Etonian Researth Counc.

Adie BFM

BPMN SIMULATION TO BE (5) (1)bpme-

BPMN Diegram wlth resuts hest mep 595 DeC Sae scmario Back to edit da

Simulation Results

General InformationComplatal procea hudances 10000 General InformationComplatal procea hudances 10000 General InformationComplatal procea hudances 10000 General InformationComplatal procea hudances 10000 General InformationComplatal procea hudances 10000 General InformationComplatal procea hudances 10000 General InformationComplatal procea hudances 10000 General InformationComplatal procea hudances 10000 General InformationComplatal procea hudances 10000
Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge Total c EURTotal siniation time 626 weeksChartsPr gof Prs F13m-31m13m-91m 1388-153m1 1000 10000Pro P-100 181Scenario StatisticsNamum Madrum Auerge
Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes Process instance cyde times induding of onetabie hours 13mnes 153mntes 19minutes
Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns Procins instance cycle times enclading off-cimdabie hours 13mm 15.3mintes 19mns
Prooes Intance coms OIUR OEUR OEUR Prooes Intance coms OIUR OEUR OEUR Prooes Intance coms OIUR OEUR OEUR Prooes Intance coms OIUR OEUR OEUR Prooes Intance coms OIUR OEUR OEUR Prooes Intance coms OIUR OEUR OEUR Prooes Intance coms OIUR OEUR OEUR Prooes Intance coms OIUR OEUR OEUR Prooes Intance coms OIUR OEUR OEUR
Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold Activity Durations,Costs, Wating times, Deviations from ThreshalsName Waiting time Duration Duration over theeshold Cest Cost over threshold
Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52 Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52 Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52 Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52 Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52 Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52 Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52 Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52 Cout Mn Ag Max Mi 서 Mox Mim R Max Ki AG Max Nom AD 52
CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01 CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01 CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01 CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01 CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01 CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01 CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01 CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01 CLCGAUDCUSTOESU PECASTRATION 204 0% 0% 27% 3% 33% 01
DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0 DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0 DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0 DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0 DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0 DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0 DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0 DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0 DAMAGE ASSESSORGDAAGEASSESSMENT 12122 05 05 30.45 33.85 3725 05 0 0
DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S DATA ANALYST& AANAOZE CLASSRICATION REPORT 295 0s 0% 0s 72% 8s 88% 0% 0S
DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175% DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175% DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175% DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175% DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175% DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175% DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175% DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175% DATA ANALYSTSCHCKREPORT QUALITY WITH RADARCAGAD 2001 144% 16% 175%
DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115 DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115 DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115 DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115 DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115 DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115 DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115 DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115 DATA ANACYSTS-AIDICONRIGURE SrSTEM 12122 05 10% 115
DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0 DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0 DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0 DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0 DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0 DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0 DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0 DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0 DATA ANALYST& A10CONFIGURE SrSTBMGa1D. 12122 0% 0% 36% 44s 0% 0% 0
DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 - DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 - DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 - DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 - DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 - DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 - DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 - DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 - DATA ANACISTS OCONTIGURE SYSTEN 12122 0% 0% 1 -
DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05 DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05 DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05 DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05 DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05 DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05 DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05 DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05 DATA ANALYST& CONFIGURE SYSTEM 12122 Os 05 1635 545 2265 05 05
DATA ANACIST&AIOSTRT SISTEM 12122 0S 18% 22% 0s DATA ANACIST&AIOSTRT SISTEM 12122 0S 18% 22% 0s DATA ANACIST&AIOSTRT SISTEM 12122 0S 18% 22% 0s DATA ANACIST&AIOSTRT SISTEM 12122 0S 18% 22% 0s DATA ANACIST&AIOSTRT SISTEM 12122 0S 18% 22% 0s 0% 0% 0% 0%
DATA ANACISTS-AIDSTAIT SISTEM 12122 0% 13% 22% 0% 13% 22% 0% 13% 22% 0% 13% 22% 0%
DATA ANACYST6410START SISTEM 12122 0s 0s 05 18% 2s 18% 2s 22s 0s 22s 0s 0s 0s 0s 0s 0s 0s 0s 0s
DATA ANALYST&45TART SISTEMSAtO 12122 18% 18% 22% 05 22% 05
MLINGMIRG ACONIGFE SISTIM 12122 0% 95% 105% 95% 105% 11.6% 01 11.6% 01 0% 0% 0% 0%
MIL BNGNEERG AIOEMALLATE AND SET NUMBEL OF GENERATION OF GRADENT PLOT 3227 05 158% 175s 158% 175s 1821 05 05 01 05 01 05 01 05 01
NL ENGINEER& AUOENALLIATE TESTING REPORT 1990 0s 0s 0% 126% 34s 1544 0s Ds 0%
MLINGNITRG ALSTART SYSTEM 12122 Os 0% O 32% 35% 32% 35% 39% 0s 39% 0s 0% 0%

5.2.3.1. Duration

5.2.3.2. Count

5.3. AS IS and TO BE MODE COMPARISON

  1. Handoff level:"Are data report balanced?" and “Is there enough data report?"were dropped since they are no more useful if report are taken from other customers of the same category.This optimization allows to save reconfigurations calls.

  2. Service level: “Clerck:definition of experts' check" has been dropped since it could be easily replaced by an automatic script that could calculate priorities and then assign damage assessors to new crash images.

  3. Task level: improvement of a ML Engineer's task : Evaluate Testing Report, where the cognitive effort of the Actor is reduced from 3 (apply) to 2 (understand), in this case the Engineer's aim is to check if the result is corrected or not, instead before that, he had to compute mathematic operations.

ML ENGINEER: EVALUATE TESTING REPORT (Enhancement)

SUB-TASK STEP ACTOR COGNITIVEEFFORT OCCURRENCE TOTALCOST
1 ML Engineerpresses the buttonto generate theevaluation testingreport. MLEngineer Remember(1) 1 13.51
2 System generates areport that containsthe top 5 NeuralNetwork withhyperparameters(Figure 1). System
3 System computesthe differencebetween errortraining andvalidation error andcompare it with athreshold. System
4 ML Engineer checksthe report's result MLEngineer Understand(2) 1 13.52
5 ML Engineerpresses the buttonto approve theclassifier. MLEngineer Remember(1) 1 13.51
6 System obtains thevalid classifiers
TOTAL= 14
Threshold 2.0
Neural Network ErrorTraining(%) ErrorValidation(%) Difference
NN_1 7.5 6.0 1.5
NN_2 8.5 8.0 0.5
2.2
NN_3 4.3 2.1 0.5
2.2
NN_4 7.9 6.0 3.9
0.1
NN_5 4.7 4.6 3.9
0.1

6. PROCESS MINING

6.1. PROM (AII)

6.1.1. Normative Model

Regarding normative model construction, multiple end events were converted into returns to the configure except for the last one ("Send label to Vehicle management system). This choice was taken because during log insertion in ProM multiple errors were generated related to the multiple endpoints. Moreover, from a logical point of view, configuration return was added to generate complete report since with endpoints only half reports were completely generated.Another option could have been to create a dummy end event and let all the end events converge to it.

Is important to highlight that our simulation on Bimp was performed using 300 tokens because some links with low probability were present so in the case of 100 tokens too many links has not been taken by any token.

6.2. DISCO vs APROMORE (AII)

6.2.1. Transiion Map before Filtering

Here the two-transition maps are compared. The two logs generated by BIMP are equal and both add two tasks (“New Report”, "Send label to Vehicle management system") more with respect to the original Normative model. To fix the transition map filters were applied to obtain

follow result.The difference between the two transition maps is the number of tokens (that we can read over the links). This is mostly caused by the fact that Apromore and Disco count tokens in two different ways. In Disco when there some loop on the number of tokens that went back and go through an activity again is added to the number of tokens who have already been there,so higher numbers w.r.t the number of tokens. In Apromore this doesn't happens, the number that can be reaf over a link is simply the number of token divided by the probability that a token go through that link. Is also highlighted that there are two loops (red circles in the Apromore's picture) and two links (red arrows in the Disco's picture) in the transition map produced with Apromore which are not present in the transition map produced with Disco.

apromore

Also, after the filtering there is some difference between the two transition maps. The same difference in the number written over the links as before (caused by the different way of counting tokens in the two software) and this time only a single loop is present (red circle in Apromore's picture) and two links (red arrows in Disco's picture) which are present in the transition map produced with Apromore but aren't in the one produced with Disco.

6.3. APROMORE STATISTICS (Domenico Armillotta, Leonardo Bellizzi)

6.3.1. APROMORE + ORIGINAL LOG FILTERED (AS IS)

Precision :0,93703

Generalization :0,99999

Precision: 0.93;

Generalization: 0,99;

Fitness: 0,97

Simplicity: 2#gateways + #sequence flows +#activities=10+36+19=65

The created model has a high fitness, so it replays the log behaviour.

It's easy to figure out, the precisionis high, so the model guarantees to allow only minimally more behaviour than seen in the logs, so there are many constraints on the behaviour.

The generalization value which means that it will be able to reproduce future behaviour of the process.

The simplicity is not very high,which means that is similar to the starting BMPN moreover it donesn't allow to have a spaghetti like model.

In the logs, the "unobservable" (grey label) correspond to split/merge events that are not Activities.

6.3.2. TO BE:

Violation List:

  • The first expert that is not working takes the job without the intervention of the Clerk.In this way "Clerk: definition of expert check” can be skipped.

  • For customers of the same category, same number of generations of the network can be used, and activity “ML Engineer: evaluate and set number of generation of the gradient plot" can be also skipped.

  • Configuration parameters set by ML Engineer (number of neurons, number of levels and threshold) are the same for customers of the same category, so no need of hyper parameterization. Under this assumption “ML Engineer: configure system" can be skipped.

6.3.3. APROMORE+MODIFIED LOG(TO BE)

ProM 6

Precision/Generalization o

Precision :0,84204

Generalization :0,99999

  • Precision: 0,84

  • Generalization: 0,99

  • Fitness: 0.97

  • Simplicity:13+19+41=73

N C

Fitness is high so the model captures log behavior in a good manner. Simplicity is average, so spaghetti like model is not present, which increases readability. Since Apromore is a mining algorithm that does not belong to the class of a algorithms, a quite higher value is expected.The precision is relatively low, so it allows different behaviors from that seen in the logs, so it does not introduce any particular constraints on behavior. Generalization is high,so the model will be able to reproduce future behavior of the process.

6.4. PROM STATISTICS (Giuseppe Aniello,Edoardo Malaspina)

Filtered logs were imported in ProM to apply two different algorithms Inductive Miner and Heuristic Miner to figure out differences between in terms of the four quality dimensions.

6.4.1. PROM INDUCTIVE MINER + FILTERED LOGS (AS IS)

  • Precision:0,93

  • Generalization: 0,99

Fitness:0,98

  • Simplicity:Σ#gateways + #sequence flows + #activitie=93519=63

The main difference with respect to the normative model is that there are less links going back to the configuration so it can be assumed that the mined model is in general good but it suffers since the simulation, which produced the logs, has been carried out with just 300 tokens so there are some branches (especially the ones going back to the configuration) that are not enough frequent to be discovered by the mining algorithm. Since there are no more tasks than expected confirms that filtering phase was well performed. The high fitness is shown by the trace fitness of the single cases in the picture above, in which there are no skipped events.The grey events are those ones are not Activities but events like “merge" or “split” due to the presence of a lot of gateways.

6.4.2. TO BE

Violation List:

  • The first expert that is not working takes the job without the intervention of the Clerk.In this way "Clerk: definition of expert check"can be skipped.

  • For customers of the same category, same number of generations of the network can be used, and activity “ML Engineer: evaluate and set number of generation of the gradient plot" can be also skipped.

  • Configuration parameters set by ML Engineer (number of neurons, number of levels and threshold) are the same for customers of the same category, so no need of hyper parameterization. Under this assumption “ML Engineer: configure system" can be skipped.

6.4.3. PROM INDUCTIVE MINER + MODIFIED LOG (TO BE)

Precision/Generalization

Precision :0,83951

Generalization :0,99999

  • Precision: 0.83

  • Generalization:0.99

  • Fitness: 0.99

Simplicity:∑Hgateways+Hsequenceflows+Hactioiiies=12+38+19=69

In this case it can figure out that the Simplicity is higher w.r.t AS IS case, this is due to the introduction of violations in the log which didn't cause the removal of any task but added some gateway and some link, for example in the left part of the BPMN model mined using the modified logs we can observe that we have one more gateway and one more link with respect to the AS IS model, in the step of the “ML Engineer:configure system" we have a new link implying the fact that sometimes that step is skipped in the logs (due to the violation introduced). The generalization in the same as before while the precision dropped because now we have more variation in the logs and given the fact that those variations are not so frequent maybe the model is not well trained on them.

7. CONFORMANCE CHECK (Normative model + Modified Logs)

As expected, when performing conformance check between modified logs and Normative Model some skipped events are encountered. In the picture above three cases are highlighted that are those in which violations in the filtered logs were inserted.

It can figure out how other cases are normal because in that case the logs haven't been modified so there are not violations, so the logs are respected in the model.

  1. OVERVIEW
MODEL FITNESS SIMPLICITY PRECISION GENERALIZATION
Original Log filtered +InductiveMiner (ProM) 0.98 63 0.93 0.99
Modified Log+Inductive Miner(ProM) 0.99 69 0.83 0.99
Original Log filtered +Apromore 0,97 65 0.93 0.99
ModifiedLog +Apromore 0.97 73 0.84 0,99

9. CLOCKIFY DASHBOARDS

9.1. Giuseppe Aniello

9.2. Domenico Armillotta

9.3. Leonardo Bellizzi

9.4. Edoardo Malaspina