This project analyses the Road Traffic Fines Management Process using process mining techniques to understand how the process is executed, identify process deviations and bottlenecks, and highlight opportunities for improvement.
The analysis was conducted using Disco for process discovery, conformance analysis, and process performance analysis, with Signavio used to create a BPMN representation of the main process flow.
The analysis focuses on four perspectives:
- Process Discovery — understanding the actual process flow and frequent process paths
- Conformance Analysis — identifying deviations in process execution
- Time Analysis — identifying delays and bottlenecks
- Resource Analysis — analysing resource workload and activity distribution
The analysis investigated the following questions:
- Does payment occur before the fine is sent?
- Are payment notifications generated after a fine has already been paid?
- What are the outcomes of appeals to the Prefecture?
- Where are the main bottlenecks in the process?
- Are activities concentrated among particular resources?
- What additional data could help explain process performance?
46,880 cases were identified where Payment occurred directly after Create Fine, indicating that payment occurred before the expected fine-notification step.
This represents a significant process deviation that warrants further investigation.
The event log contained:
- 561,470 events
- 150,370 cases
- Median case duration: 28.3 weeks
- Mean case duration: 48.8 weeks
The higher mean compared with the median indicates the presence of cases with very long durations.
Send for Credit Collection occurred in 59,013 cases and was identified as a major area for further investigation based on its frequency and process duration.
The dataset contained 149 resources.
Resource 538 recorded 64,097 occurrences, representing approximately 29.81% of the events handled by the identified resources.
This indicates significant workload concentration around a single resource and provides an area for further investigation into resource allocation and process performance.
Among 532 analysed cases involving an appeal to the Prefecture:
- 314 cases resulted in fines being reassessed and subsequently paid
- 218 cases resulted in cancelled fines but were subsequently routed toward credit collection
The 218 cancelled cases represent approximately 41% of the analysed appeal cases and highlight a potential process-handling inconsistency that warrants further investigation.
The event log was loaded into Disco to identify the actual process flow and the most frequent paths.
The initial process map contained a large number of activities and paths, so filtering was applied to make the dominant process behaviour easier to analyse.
Phase 1
- Activities: approximately 70%
- Paths: approximately 50%
Phase 2
- Activities: approximately 30%
- Paths: approximately 10%
- Frequency filtering was applied to focus on the most frequent variants
The process was then represented using a BPMN model in Signavio to provide a clearer representation of the main process flow.
The main activities represented in the BPMN model include:
Create Fine
↓
Send Fine
↓
Insert Fine Notification
↓
Appeal / Payment
↓
Add Penalty
↓
Payment
↓
Send for Credit Collection
The appeal process includes activities related to sending an appeal to the Prefecture, receiving the result, and notifying the offender.
Rare exceptions were excluded from the simplified BPMN representation to keep the model focused on the dominant process behaviour.
Three process behaviours were investigated using filters in Disco.
The sequence:
Create Fine → Payment
was investigated using a follower filter.
Result: 46,880 cases.
This indicates a significant deviation from the expected process sequence and warrants further investigation into the underlying process or event-log behaviour.
The sequence:
Payment → Insert Fine Notification
was investigated.
Result: 74 cases.
These cases indicate that a fine notification may still be generated after payment has already occurred.
The sequence involving:
Receive Result Appeal from Prefecture → Notify Result Appeal to Offender → Payment
was investigated.
Among 532 cases:
- 314 cases involved reassessed fines followed by payment
- 218 cases involved cancelled fines and subsequent routing toward credit collection
This highlights a potential inconsistency in the handling of cancelled appeal outcomes.
The process statistics showed:
- Median case duration: 28.3 weeks
- Mean case duration: 48.8 weeks
The difference between the mean and median indicates that some cases have substantially longer durations.
Two activities with particularly long durations identified in the analysis were:
- Insert Date Appeal to Prefecture: 22.6 months for 39% of cases
- Appeal to Judge: 30.6 months for approximately 0.1% of cases
Although the Appeal to Judge activity has a very long duration, its low frequency means it represents a rare exception.
The dataset contains 149 resources.
Resource analysis showed that Resource 538 had the highest frequency, with 64,097 occurrences, representing approximately 29.81% according to the resource analysis in the project report.
Other resources handled substantially fewer events.
This concentration suggests a potential workload imbalance and provides an area for further investigation into:
- Resource allocation
- Workload distribution
- Processing delays
- Resource performance
Additional resource-level information would be required to determine whether workload concentration directly contributes to process delays.
No missing or invalid data was identified during the analysis.
The project examined both short-duration and long-duration cases to understand whether data completeness could explain process delays.
The analysis found that long-duration cases appeared to contain complete event information, suggesting that the observed delays may require investigation from process, resource, or system perspectives rather than being attributed directly to missing data.
Additional information that could provide deeper insights includes:
- Detailed resource performance data
- Exception handling information
- Customer feedback and satisfaction data
- Event-level comments and reasons for actions
- Payment method information
Based on the process-mining analysis, the following areas were identified for further investigation.
Review the 46,880 cases where payment occurred directly after creation of the fine to determine whether these represent legitimate exceptions, system behaviour, or event-log sequencing issues.
Analyse the activities and waiting times surrounding Send for Credit Collection to identify potential sources of delay.
Investigate the workload concentration around Resource 538 and assess whether responsibilities could be distributed more effectively.
Investigate why cases with cancelled fines subsequently appear in the credit-collection flow.
Although no missing or invalid data was identified in the analysed event log, automated validation at the point of data entry could help maintain data quality.
The repository contains the main visualizations used during the analysis.
- 📈 Disco Process Map — Process discovery and major process paths
- 🔎 Disco Conformance Analysis — Detailed process filtering and conformance analysis
- 🔄 BPMN Process Model — Simplified BPMN representation of the process
| Tool / Technology | Purpose |
|---|---|
| Disco | Process discovery, filtering, conformance and performance analysis |
| Signavio | BPMN process modelling |
| XES | Event-log format |
| BPMN 2.0 | Process representation |
| Detail | Information |
|---|---|
| Dataset | Road Traffic Fines Management Process |
| Type | Real-world event log (.xes) |
| Source | 4TU Research Data |
| Total Events | 561,470 |
| Total Cases | 150,370 |
| Activities | 11 |
| Time Period | 01/01/2000 — 06/18/2013 |
process-mining-analysis/
├── bpmn-process-model.jpeg
├── disco-process-map.jpeg
├── disco-conformance-analysis.jpeg
└── README.md