Process Mining - Structure of transformations

Overview

Below is an overview of the transformation steps of the Process Mining app templates.

The models\ folder is organized according the structure of the transformation steps.

1. Input

The input step is used to load the raw data. The following operations are typically done to prepare the data for the next transformation steps:

Select fields with the optional and mandatory macro. A field doesn't need to be present in the raw data when the optional macro is used.
Type cast fields to the appropriate data types.
Filter tables to reduce data size early in the transformations.

Note: It is recommended to reduce data size already in the extraction where possible.

Naming convention

If you expect name clashes with table names in next transformation steps, it is best practice to add the suffix _input to the input tables.

Optional Fields

The input of a connector consists of mandatory and optional data. For optional fields, the optional macro from the pm-utils package should be used. This makes sure that a field gets the value null if it is not available in the source data. That way, all transformations will run correctly.

Use the optional_table() macro if the whole table is optional.

2. Entities

In the entities step, input tables are transformed to entity tables. Each entity required for the expected events should get its own table. Refer to Designing an event log. Additionally, supporting entities can also be defined here.

In the below example 3 input tables Invoices_input, Invoice_types_input, and Customers_input are joined together to create the entity table Invoices.

Guidelines

Follow these guideline when creating an entity table.

There is one entity ID field, which is unique for each data record.
All entity fields that are needed for data analysis are present.
All entity fields have names that are easy to understand.

When applicable, the entity table relates to another entity via an ID field. Refer to the example below, where the invoice lines are related to the invoice entity via the Invoice_ID field.

Additional transformations

Not all input tables are transformed into entity tables. Also, other input tables may contain relevant information, such as the Customers table in the example. It may be convenient to define them in the entities step as separate tables such that they can be reused in the data transformations.

Naming convention

If the entity table names would lead to name clashes later on, add the suffix _base to the tables.

3. Events

Note: The input for Event log and Custom process app templates is already a well defined event log for Process Mining. There is no need to transform the data from the source system into the events for Process Mining here. This means that the 3. events is not present in the transformations for Event log and Custom process apps.

In this transformation step, event tables are created for each entity. See Designing an event log. Each record in an event table represents one event that took place. There are two scenarios on how the data is structured:

Timestamp fields: Fields on an entity table with a timestamp for an event. For example, the Invoice_created field in an Invoices table.
Transaction log: A list of events.

Based on how the data is structured, the transformations to create the event tables are different.

Timestamp fields

In this scenario, the values of a timestamp field must be transformed into separate records in an event table. The below example is an invoices table that contains three timestamp fields.

Each timestamp field is used to create a separate event table. For every record that the timestamp field contains a value, create a table with the Invoice ID, the name of the event (Activity), and the timestamp the event took place (Event end).

The Invoices_input table is split into Invoice_events_Create_invoice, Invoice_events_Delete_invoice, and Invoices_events_Change_invoice_price.

The separate event tables can then be merged into a single event table per entity, for example Invoices_events.

Transaction log

If events are stored in a transaction log, the relevant events per entity should be identified. Create a table per entity and store corresponding entity ID, the name of the event (Activity), and the timestamp the event took place (Event end).

In the below example, the transaction log contains events for the Purchase Order and Invoice entities.

The following fields are mandatory in an event table. All records in the event tables should contain a value for these fields.

Field	Description
Entity ID	ID of the entity for which the event happens. For example, the Invoice ID.
Activity	The activity describes which action took place on the entity.
Event end	The event end field indicates when the specific event was finished. Ideally, this should be a datetime field, rather than a date.

Naming convention

Name the tables according to the structure [Entity] + _events. For example, Purchase_order_events and Invoice_events.

4. Event logs

Single entity process

When the process contains one entity, no additional transformations are needed in this step. The single entity table and events tables are already in the correct format.

Multiple entity process

When multiple entities are involved in a process, the events of all entities need to be linked to the main entity that is considered the “Case” in the process. Refer to Define the event log for details. The below steps describe how to relate all events to the main entity, and how to combine them a single event log.

Entity relations

Create an “entity-relations” table to centralize the relationships between all entities. This entity-relations table will contain the ID fields of the related entities.

To create the entity-relations table, join all entity tables based on their ID fields:

Start with the main entity
Join related entities to the main entity with a left join.
If entities do not relate directly to the main entity, left join them to the related entities that are already joined to the main entity.

In the below example, there are three entities: Purchase order,Invoice line, and Invoice. The Purchase order is considered the main entity in the process. The Invoice line is directly linked to the Purchase order and the Invoice is linked indirectly via the Invoice line.

Below is the resulting entity-relations table.

Relation tables

The individual relations between the main entity and each other entity are stored in separate tables, using the combined information from the entity relations table.

Event log

The next step is to use these relations to add corresponding “Case ID” to each event table. The "Case ID" is obtained via the relation table, where the event information is obtained from the event table. To create the full event log, the event tables for each entity are unioned.

Naming convention

If the event log table name can lead to name clashes at a later stage, add the suffix  _base to the name of the event log tables.

5. Business logic

In the last transformation step, business logic is added as needed for data analysis. Additional derived fields can be added to existing tables here. For example, specific throughput times or Boolean fields that are used in KPIs in dashboards.

In Process Mining, there are two additional standard tables defined in this transformation step: Tags and Due dates.

Due dates

Due dates represent deadlines in the process. These are added to the data to analyze whether activities are performed on time for these due dates or not.

Each record in the due dates table represents one due date for a certain event. Example due dates are:

a payment deadline for a payment event.
an approval deadline for an approval event.

The mandatory fields for this table are the Event ID, Due date, Actual date, and Expected date.

Not all events will have a due date and some events may have multiple due dates.

Structure of transformations

Overview

1. Input

Naming convention

Optional Fields

2. Entities

Guidelines

Additional transformations

Naming convention

3. Events

Timestamp fields

Transaction log

Naming convention

4. Event logs

Single entity process

Multiple entity process

Entity relations

Relation tables

Event log

Naming convention

5. Business logic

Tags

Due dates

Was this page helpful?