MVMTN

MVMTN: A multi-task learning framework for multi-view predictive business process monitoring

Requirements

python==3.9.12
Levenshtein==0.21.0
matplotlib==3.5.1
numpy==1.22.3
pandas==1.5.2
pandoc==2.3
pm4py==2.7.4
sklearn==1.1.3
torch==1.12.0
torch-geometric==2.1.0
tqdm==4.64.1

Prepare

raw datasets are placed in path 'raw_datasets'

use 'mkdir datasets_results' and 'mkdir processed_datasets' to store the datasets.

run 0.raw_dataset_process.ipynb to process xes format datasets into csv format.

Pattern mining

run 1.multi_view_pattern_mining.ipynb to mine frequent pattern from different view

Trace matching

run 2.prefix_generating_and_trace_matching.ipynb to perform prefix generating and trace matching

Multi-task learning

use 'mkdir results' to store the results of MTL model

run 3.multi_task_learning.ipynb to train the MTL model

run 3.multi_task_learning_no_frequent.ipynb to train the model without frequent learning module

run 3multi_task_learning_no_task.ipynb to train the model without task aggregation module