Predicting Activities in Business Processes with LSTM Recurrent Neural Networks

Published: 26 June 2018| Version 1 | DOI: 10.17632/trskzyg3j9.1
Contributors:
Jorge Roa,
Mariano Rubiolo,
Edgar Tello-Leal

Description

The Long Short-Term Memory (LSTM) Recurrent Neural Networks provide a high precision in the prediction of the modeling of sequences in several application domains. This article introduces the use of LSTM networks for the prediction of activities in a business process. This is a key aspect to provide valuable input for planning and resource allocation. Predicting the behavior of a business process is possible by exploiting event logs to make predictions about execution of cases. Each trace associated with a case indicate the sequential execution of activities. A methodology for the implementation of the LSTM network in the process mining domain is also proposed. An event log of the industry domain is used to train and test the proposed LSTM neural network. Our preliminary results indicate that the prediction of the next activity is acceptable according to the literature of the domain.

Files

Steps to reproduce

SYSTEM REQUIREMENTS ** python version 3.6 or newer ** tensorflow (pip install tensorflow) ** keras (pip install keras) ** h5py (pip install h5py) REPRODUCING THE PAPER RESULTS ** RUN lstm_predict_kaleidoscope.py GENERATE NEW MODEL ** EDIT lstm_generate_kaleidoscope.py and set the new parameters ** SAVE the file ** RUN lstm_generate_kaleidoscope.py

Institutions

Universidad Autonoma de Tamaulipas, Universidad Tecnologica Nacional Facultad Regional Santa Fe

Categories

Business Process Management, Machine Learning, Mining

License