Page 107 - Proceedings of the 2018 ITU Kaleidoscope

Predicting Activities in Business Processes with LSTM Recurrent Neural Networks

1
2,3
2
Edgar Tello-Leal ; Jorge Roa ; Mariano Rubiolo ; Ulises M. Ramirez-Alcocer 1
1
Faculty of Engineering and Science, Autonomous University of Tamaulipas, Victoria, Tamaulipas, Mexico, 87000
2
CIDISI, Santa Fe Regional Faculty, National Technological University, Lavaisse 610, (S3004EWB) Santa Fe, Argentina.
3
Research Institute for Signals, Systems and Computational Intelligence, sinc(i), FICH/UNL-CONICET, Ciudad
Universitaria, (3000) Santa Fe, Argentina.

ABSTRACT number of activities that organizations perform are supported
by information systems. Several types of activities contained
The Long Short-Term Memory (LSTM) Recurrent Neural in the business processes can be executed automatically by the
Networks provide a high precision in the prediction of information systems, without the participation of a human.
sequences in several application domains. In the domain
The convergence of solutions and products towards the
of business processes it is currently possible to exploit event
BPM and the Service Oriented Architecture (SOA) paradigm
logs to make predictions about the execution of cases. This
adopted for industrial systems contributes to the improvement
article shows that LSTM networks can also be used for the
in the reactivity and performance of industrial processes
prediction of execution of cases in the context of an event
such as manufacturing or logistics among others [3]. This
log that originates from the IoT and Industry 4.0 domain.
is leading to a situation where information is registered in
This is a key aspect to provide valuable input for planning
event logs, making it available in near-real time based on
and resource allocation (either physical or virtual), since
asynchronous events, and to business-level applications that
each trace associated with a case indicates the sequential
are able to use high-level information for various purposes,
execution of activities in business processes. A methodology
such as diagnosis, performance indicators, or traceability. [3].
for the implementation of an LSTM neural network is also
In this context, predicting the behavior of a business process,
proposed. An event log of the industry domain is used to train
i.e. exploiting event logs to make predictions about the
and test the proposed LSTM neural network. Our preliminary
execution of activities [4], is a key aspect in order to provide
results indicate that the prediction of the next activity is
valuable input for planning and resource allocation [4]. There
acceptable according to the literature of the domain.
are two main factors for the growing interest in predicting
business process behavior. On the one hand, with the advent
Keywords - LSTM, event log, process mining, business
of 5G and the Internet of Things (IoT) in the context of
process
Industry 4.0 [3], more and more events are recorded due
to the great number of devices connected to the Internet,
1. INTRODUCTION
providing detailed information about the history of business
In a knowledge-based economy, public and private processes. On the other hand, there is a need to improve
organizations require proper knowledge asset management and support business processes in competitive and rapidly
to maintain a competitive advantage in global markets or changing environments.
in government services. With the advent of robotics, Process mining techniques [5, 6] are capable of extracting
machine learning, and 5G networks there will be a wealth knowledge from event logs, commonly available in
of opportunities for cooperation between robots and humans information systems. These techniques provide new means to
improving productivity and speeding up the delivery of discover, monitor and improve business processes in a variety
services for citizens. In this context, the Business Process of application domains. However, standard process mining
Management (BPM) is considered a key component to techniques cannot deal with predicting process behavior.
managing the life-cycle of business processes that orchestrate The recurrent neural network (RNN) architecture has become
the activities performed in organizations as well as the a model of the neural network implemented in different
resources (humans, robots, or information systems) that domains, due to its natural ability to process sequential entries
execute such activities. and to know their long-term dependencies [7]. Unlike the
A business process consists of a set of activities that are feed-forward neural network, the RNN neurons are connected
performed in a coordinated way in an organization in a to each other in the same hidden layer and a training function
technical environment, and have at least one correlated is applied to the hidden states repeatedly [7]. The Long
business goal [1]. The standard language for modeling Short-Term Memory (LSTM) neural network is an extension
business processes is BPMN (Business Process Modeling of the RNN, which has achieved excellent performance in
Notation) [2]. Information technologies in general and various tasks, especially for sequential problems [8], [9],
information systems in particular play an important role [10]. The implementation of LSTM neural networks for the
in the management of business processes, because a large discovery of events or activities of a business process through

978-92-61-26921-0/CFP1868P-ART @ 2018 ITU – 91 – Kaleidoscope