Page 108 - Proceedings of the 2018 ITU Kaleidoscope

‎ 2018 ITU Kaleidoscope Academic Conference‎

predictive analysis can be considered an important strategy as discovered is typically a business process model represented
a technique of process mining and has been used with success using a graphical notation such as the BPMN language [2],
in this domain [4, 11]. Petri nets [13, 14, 15], Event-driven Process Chains (EPC)
In this work, we propose an approach for the discovery of [16], or UML activity diagrams [17]. Process conformance
events and activities of a business process through predictive consists of comparing a business process model with the
analysis from traces contained in event logs taken from the event record generated by the execution of the same process
IoT and Industry 4.0 domain. The predictive model is based model [6]. Conformance verification can be used to evaluate
on an LSTM recurrent neural network that is trained with whether the information stored in the event log is equivalent
event logs, enabling the prediction of the following activity to the model and vice versa. Process improvement consists of
in a trace of execution that follows another activity or a set extending or improving an existing process model using the
of activities given as input. In order to validate the approach stored information of the current process in the event log.
and show the applicability to the proposed domain we present
preliminary results based on a dataset with 255 traces. The 2.2 LSTM Neural Networks
test carried out on the trained LSTM network shows that it has
the capacity to predict the next activity of a business process Recurrent neural networks (RNN) with Long Short-Term
model. Memory (LSTM) emerged as an effective and scalable model
This work is structured as follows. Section 2 presents the for learning problems related to sequential data [18]. RNNs
background. Section 3 presents an introductory example. have two types of input, the present, and the recent past. RNN
Section 4 introduces an approach to predict business process use both types of input to determine how they behave with
behavior. Section 5 shows the results. Section 6 presents respect to new data. This means that the output of a RNN
related work. Finally, Section 7 concludes this work and at time step t-1 affects its output at time step t. LSTMs
proposes future work. are general and effective at capturing longterm temporal
dependencies [18].
2. BACKGROUND The information contained in LSTMs are outside the normal
flow of the recurrent network in a gated cell. Information
2.1 Process Mining can be stored, written or read from a cell, similar to data
in a computer’s memory. The cell makes decisions about
Process mining is an area of research that is located, on the one what should be stored and when it should be allowed to read,
hand, between computational intelligence and data mining, write and delete, through gates that open and close. These
and on the other hand, between business process modeling gates are implemented with the multiplication of elements by
and analysis. There are several areas that are included sigmoids, which are all in the range of 0-1.
in process mining, such as process discovery, compliance
verification, process improvement, organizational mining, 3. INTRODUCTORY EXAMPLE
process model extension, automatic repair of process models,
case prediction, automatic construction of models based on Industries work to increase the overall effectiveness of
simulation, and recommendations based on the history of their plants and equipments, in order to get better
execution of processes. system integration, availability, maintainability, performance,
In process mining, it is assumed that it is possible to record quality, or functionality. [3]. The example analyzed in
events sequentially since each event has a reference to an this paper is based on [19] and focuses on the control of a
activity and is related to a particular case (an instance of plant to increase its overall performance, including predictive
the process) [6]. Then, the input data in the process mining maintenance. The plant produces parts made of metal such
is an event log. An event log is a hierarchically structured as spurs, fastener, ball nuts, discs, tubes, wheel shafts, or
file with data on the executions of business processes [12]. clamps. To build these parts there are 28 machines for
This file contains data on several executions of the same lapping, milling, turning, sinking, wire cutting, turning and
business process. An event is the atomic part of the execution milling, laser marking, and round and flat grinding.
of a specific process and may contain a large number of Table 1 shows an excerpt of the event log of the process
attributes. Event data, generated by information systems, is that controls the logic of the plant. The complete log can
usually found as updates to a state (for example, the status of be found in [19]. The dataset contains process data from
"sent invoice" changes to the status "paid invoice"), or also a production process, including data on cases, activities,
as activity records (for example, " email sent to the client "). resources, timestamps, and more data fields.
A trace is a set of events that belong to the same execution It is known that, in general, with higher quality and
of a business process. Therefore, event logs can contain information coming from sensors in the process and the
additional information about events, such as the user who critical equipment for the control of the plant, it is possible
runs the activity or device that initiates the activity, the time to improve the plant operation and production planning [3].
the event started, the duration of the event, among others. In this scenario, exploiting event logs, that provide detailed
The main tasks of process mining are discovery, compliance information about the history of business processes and
and process improvement [6]. Process discovery consists of register sensor information, for predicting the next activity
using an event log as input and producing a business process to be executed in a business process is important to provide
model without using a-priori information [6]. The model valuable input for planning and resource allocation, such as

– 92 –