Process mining

Process mining is a data mining technique based on the analysis of log files. As a method of process man­age­ment, process mining offers the pos­sib­il­ity of analysing business processes and identi­fy­ing potential for op­tim­isa­tion.

Process mining comprises tech­niques in the area of business process man­age­ment that serve to analyse business processes. There are data-supported methods of process analysis that focus on the eval­u­ation of event logs – in­form­a­tion stored in IT systems about in­di­vidu­al process steps. Process mining ap­plic­a­tions apply special data mining al­gorithms to log files and trans­ac­tion data to identify trends and patterns. The aim is to gain a better un­der­stand­ing of relevant business processes in order to make them more efficient.

In research, process mining is also referred to as “Automated Business Process Discovery” (ABPD) and describes tech­niques used to create, evaluate, and extend process models. The Process Mining Manifesto by the IEEE Task Force on Process Mining dis­tin­guishes between three types of process mining tech­niques:

• Discovery: Discovery process mining tech­niques are used to identify processes and create process models.
• Con­form­ance: Con­form­ance process mining tech­niques enable an as­sess­ment of the con­form­ity of existing process models to current data.
• Extension/en­hance­ment: Extension (also known as en­hance­ment) process mining tech­niques are used to enhance existing process models.

Process mining combines data mining and com­pu­ta­tion­al in­tel­li­gence (CI) tech­niques with process modelling and analysis. A process is described as a series of logically linked process steps that can be recorded as events.

The starting point for any process mining technique is event data in the form of log files that reproduce events in chro­no­lo­gic­al order and can be assigned to both a process step and a process instance.

The result of this type of process mining is usually a process model. In a man­u­fac­tur­ing plant, for example, a model like this could be derived from time stamps that indicate when which product passes through which pro­duc­tion step.

Common present­a­tion tech­niques for process models are:

• BPMN (Business Process Model and Notation)
• EPC (Event-driven Process Chain)
• HIPO diagrams
• Com­mu­nic­a­tion structure analysis
• Petri net models
• SOM (Semantic Object Model)
• UML (Unified Modeling Language)
• BPEL (WS-Business Process Execution Language)

“Con­form­ance” process mining tech­niques are used to validate process models. If a process model already exists, it is advisable to compare it at regular intervals with new event log data to ensure that the model cor­res­ponds to how the real processes are being doc­u­mented. Process mining tech­niques are used to compare the existing process model with current event data in order to determine dif­fer­ences between the model and reality. The resulting diagnosis of a con­form­ance test like this enables con­clu­sions to be drawn about the quality of the process model under in­vest­ig­a­tion. A con­form­ance test can be applied to both de­script­ive and normative process models.

Process mining covers four different levels of ob­ser­va­tion:

• Control flow per­spect­ive: A process with a view to the control flow aims to represent the sequence of activ­it­ies within a process as a process model (e.g. as a petri net UML activity diagram, EPC, or BPMN model.
• Or­gan­isa­tion­al per­spect­ive: Process mining from an or­gan­isa­tion­al per­spect­ive high­lights how people and IT systems relate to each other through par­ti­cip­a­tion in a business process. Activity profiles and roles are defined and compared with each other. The result of an analysis like this is a social network that visu­al­ises the network of re­la­tion­ships.
• Case per­spect­ive: Process mining with a case per­spect­ive is used to analyse in­di­vidu­al process instances. These are described and cat­egor­ised as cases according to their prop­er­ties. The clas­si­fic­a­tion takes place according to the data values recorded for the re­spect­ive process instance – for example, according to which actors are involved.
• Time per­spect­ive: Process mining with a time per­spect­ive takes a close look at the absolute or relative point in time and the frequency of events. The pre­requis­ite for this is that all event logs have a time stamp. Analyses of this kind allow sim­u­la­tions that enable con­clu­sions to be drawn about patterns, trends, and obstacles in the process flow. For example, bot­tle­necks in the process chain can be iden­ti­fied.

In practice, process mining today is primarily used for control flow detection. In the fore­ground are the “discovery” process mining tech­niques with a control flow per­spect­ive, which make it possible to identify the chro­no­lo­gic­al sequence of in­di­vidu­al process steps and to compare them with the desired target state.

The IEEE has developed the L* life-cycle model as a reference model for applying process mining tech­niques. This divides the procedure for process mining projects into five phases:

Process mining can be used wherever detailed in­form­a­tion about the in­di­vidu­al steps of relevant business processes is recorded and per­man­ently stored with the help of IT systems. It can be used, for example, when companies:

• Process workflows via workflow man­age­ment systems
• Make trans­ac­tions using ERP systems
• Manage support requests via a ticket system
• Ensure the quality of medical treatment via clinical treatment pathways

This makes process mining suitable for use in retail and OEM, banking, de­vel­op­ment, sales, and the insurance industry to improve business processes such as ordering processes, man­u­fac­tur­ing processes, or cash flows.

Workflow man­age­ment and knowledge man­age­ment are central fields of ap­plic­a­tions for process mining tech­niques. In addition, knowledge gained from process mining projects is used in the de­vel­op­ment of as­sist­ance systems.

Many companies use tech­no­lo­gies such as databases, ERP systems, and knowledge man­age­ment systems to safeguard factual knowledge. As a rule, process knowledge is not processed. This is where process mining comes in with methods, which make it possible to make implicit process knowledge explicit.

Workflow man­age­ment systems describe business processes at formal levels and automate the co­ordin­a­tion and control of in­di­vidu­al process steps. The system provides users with user in­ter­faces for com­mu­nic­a­tion and for accessing data and programs. Workflow man­age­ment is based on modeled workflows that allow the system to recognise events (such as inputting a document by e-mail) and auto­mat­ic­ally react to them. This auto­ma­tion is based on process models that can be created, checked, and extended using process mining methods.

Process mining tech­niques can be used wherever in­di­vidu­al steps of business-relevant processes are recorded as logs. Al­gorithms from the fields of data mining and com­pu­ta­tion­al in­tel­li­gence now make it possible to analyse even complex event data and derive insights into how business processes can be made more efficient and secure.

The high degree of auto­ma­tion dis­tin­guishes process mining from classical tech­niques for creating process models. By ex­tract­ing in­form­a­tion on real events from the operative business, process mining methods real­ist­ic­ally reproduce process sequences. Compared to manual tech­niques, process mining scores points when it comes to speed and accuracy. In addition, the in­creas­ing volume of data can already no longer be managed manually.

Another advantage of pro­fes­sion­al process mining ap­plic­a­tions are the extensive visu­al­isa­tion options. Process models are presented to skilled workers and managers on in­ter­act­ive dash­boards, which enable a dynamic view of process flows and sometimes provide ad­di­tion­al analysis tools.

Companies encounter dif­fi­culties when im­ple­ment­ing process mining tech­niques when the data basis to be analysed is in­con­sist­ent due to a het­ero­gen­eous IT in­fra­struc­ture. If uniform de­scrip­tions for events are missing, the cor­res­pond­ing log files must first be processed. This not only means ad­di­tion­al effort, but may also result in data cor­rup­tion – which, strictly speaking, no longer rep­res­ents real data.

In addition, companies are con­fron­ted with technical hurdles during im­ple­ment­a­tion. The use of data mining is only effective if the re­spect­ive ap­plic­a­tions have access to all relevant IT systems. This requires ap­pro­pri­ate in­ter­faces and for the connected systems to be con­figured properly, as well as close co­oper­a­tion with the provider of the process mining ap­plic­a­tion.

The effort required for im­ple­ment­a­tion also increases when companies combine standard ap­plic­a­tions for managing business processes with tools that they have developed them­selves in order to adapt them to in­di­vidu­al needs.