Business Process Intelligence 2019

1 Preprocessing

name	description	unique values	type
eventID [Event]	identifier for events	1595923	int64
case Spend area text	type/class of purchase item	21	string
case Company	company from which purchase originated	4	string
case Document Type	type of order in the case	3	string
case Sub spend area text	type of purchase item	136	string
case Purchasing Document [PO]	identifier for cases/purchases	76349	int64
case Purch. Doc. Category name	category of purchase document	1	string
case Vendor	vendor to which purchase document is sent	1975	string
case Item Type	purchase classification	6	string
case Item Category	purchase document matching category	4	string
case Spend classification text	purchase type classification	3	string
case Source	system from which purchase originated	1	string
case Name	name/vendor of purchase	1899	string
case GR-Based Inv. Verif.	case Goods Received invoice verification	2	bool
case Item	item type	490	int64
case concept:name [Case]	case identifier	251734	string
case Goods Receipt	flag indicating if 3 way matching is required	2	bool
event User	user identifier of the event	628	string
event org:resource	DUPLICATE user identifier of the event	628	string
event concept:name	event name/type	42	string
event Cumulative net worth (EUR)	cost of purchase at the time of the event	25221	float64
event time:timestamp	timestamp of the event	167432	datetime

Table 1 - Variable Specification

1.1 Missing Values

• 3289 cases. with missing values in 16294 events.
• Missing variables:
  • case Spend area text
  • case Sub spend area text
  • case Spend classification text
• For each case with missing values, each activity in this case is missing values for these attributes.
• These missing values are filled with value Missing
  • Cases are not removed, as the fact that these values are missing may not indicate faulty data, but intended design as part of the process. (e.g. particular Spend classification does not fall under the available categories, so is left empty.)
  • These variables missing from a case may be useful in (causal) analysis of process deviations.

1.2 Timestamps

• Since the BPI Challenge Description specifies that the date is from purchase orders submitted in 2018, we analyze the events, cases and purchase documents occuring before and after 2018.
• The events prior to 2018 have an interesting distribution of events (Table 2), only 3 types of events out of the 42 occur, and they all occur during midnight.

event concept:name	count	unique	top	freq
Create Purchase Requisition Item	7	1	00:00:00	7
Vendor creates debit memo	27	1	23:59:00	27
Vendor creates invoice	284	1	23:59:00	284

Table 2 - Events occuring before 2018


Furthermore, we look at the amount of purchasing documents starting and ending each year, and see that there is a significant number of purchasing documents starting or ending after 2018 (Table 3). Since they can be scheduled events, we decided to keep all cases and purchasing orders that do not have any events prior to 2018. The other cases and thus purchasing documents are dropped.

Year	Events	Purchasing Documents starting	Purchasing Documents ending
1948	10	1	0
1993	9	1	0
2001	22	15	0
2008	45	1	0
2015	3	2	0
2016	6	2	0
2017	223	75	0
2018	1550468	76241	65268
2019	45135	11	11079
2020	2	0	2

Table 3 - Purchasing Documents and Events with Purchasing Order start and end times

Additional Findings for 2018

• Peak in events during day 27-30 of the month
• Peak in events during 00h-02h, 08h-15h and 21h-24h of the day.
• Peak in events during January, December in a year.

1.3 Redundant columns

• The columns case Source and case Purch. Doc. Category name are dropped as they contain only 1 value for the entire dataset. (Table 1)
• The column event org:resource is dropped as it is duplicate to event User

1.4 Value Counts and Distributions

• case Company is almost always Company0000 (99.63%).
• case Document Type is almost always Document Standard PO (96.49%)
• case Goods Receipt is almost True (99.63%), this means that the third category (2-way matching) almost never occurs.

2 Entity Relationships

Which of the variables is our case key?

• case Vendor has multiple case Name's, use case Vendor as primary key for case Name

• case concept:name which corresponds to the Purchase item

• UML diagram was made

• What entities do we have? Which attributes are specific to Purchase Document?

• "case Spend classification text" is interesting: it is not a Document Variable, since one Purchasing Document does not only have one value for this attribute, but most of Purchase Docs indeed only have one, only very few(2%) have multiple(exactly 2) values of this Hypothesis: These are empty values or other values and should be filled in with the most occurring value for a Doc

• "case Sub spend area text" is unique for 71469 Purchase Documents, the rest has multiple values. We therefore keep it as purchase item attribute, although it is greatly influenced by which purchase doc does the item belong to

• Most purchase documents only have one single purchase item (in fact, the median amount of purchase items per purchase document is 1, 75 percentile is 2)

• When looking at the 0.1% of purchase documents with the most purchase items, still 75th percentile is only 131, while maximum amount of purchase items per purchase document is 429.

• Surprisingly, "case Goods Receipt" is a variable of the Purchasing Document, although GR-based

Variable	Purchasing Document (max unique)	case concept:name (max unique)	Choice
case Goods Receipt	1	1	Purchasing Document
case Company	1	1	Purchasing Document
case Document Type	1	1	Purchasing Document
case Purch. Doc. Category name	1	1	Purchasing Document
case Vendor	1	1	Purchasing Document
case Name	1	1	Purchasing Document
case Source	1	1	Purchasing Document
case Spend classification text	2	1	case concept:name
case GR-Based Inv. Verif.	2	1	case concept:name
case Item Type	2	1	case concept:name
case Item Category	3	1	case concept:name
case Spend area text	4	1	case concept:name
case Sub spend area text	7	1	case concept:name
case Item	429	1	case concept:name

3 Petri Nets & Process Models

4 Data Analysis