Incremental predictive monitoring of Business Processes with a-priori knowledge

Description

Continuation of this project in which a predictive model is tasked to predict the continuation of a business process.

The input consists of categorical variables, indicating an activity ID and a resource ID, as well as real-valued variables, indicating for example elapsed time. The prediction consists of next time-step variables.

In this project, we leverage given Multi-Perspective A-priori Knowledge to improve inference on new data.

This repo is based on code from:

• Process-Sequence-Prediction-with-A-priori-knowledge
• ProcessSequencePrediction

The LTLCheckForTraces.jar program is an artifact generated from the code at this repo.

Predictive model

This contribution aims at improving the existing predictive model only, without improving/developing the existing inference algorithms used for evaluation.

Inference algorithms

The project is divided into Control Flow (CF) prediction and Control Flow + Resource (CFR) prediction. At the moment, control flow and resource consist of categorical variables. Time has also been added as a prediction element, creating the Control Flow + Resource + Time (CFRT) acronym.

Control Flow inference algorithms

• baseline_1_cf -> Baseline 1 - no a-priori knowledge is used and only the control-flow is predicted.
• baseline_2_cf -> This is Baseline 2 - a-priori knowledge is used on the control-flow and only the control-flow is predicted.

Control Flow + Resource inference algorithms

• baseline_1_cfr -> Extended version of Baseline 1, where also the resource attribute is predicted.
• baseline_2_cfr -> Extended version of Baseline 2, where a-priori knowledge is used on the control-flow but also the resource attribute is predicted.
• new_method_cfr -> Proposed approach, where a-priori knowledge is used on the control-flow and on the resource attribute. Both the control-flow and the resource are predicted.

Project structure

The src folder contains all the scripts used.

experiment_runner is used to train and evaluate the predictive models.

The inference_algorithms folder contains all the inference algorithms available.

Results are saved into the output_files folder.

shared_variables.py contains meta-variables used at inference-time, as well as training/model hyperparameters.

parse_results.py parses the results contained in the results folder inside each model run, and plots multiple images/tables that allow quick comparisons.

Getting started

This project is intended to be self-contained, so no extra files are required.

Prerequisites

To install the python environment for this project, refer to the Pipenv setup guide

Running the algorithms

As the Java server is now integrated in the project, there is no need to start it separately.

The experiment_runner.py file contains all the code necessary to train the predictive models on each log file and evaluate them with each inference method.

The run_experiments script allows the user to train both old and new implementations of the predictive model on all the datasets automatically.

Training the predictive models

Simply run the experiment_runner.py script with the --train flag. If no option is specified the script will train and test automatically the model

Evaluating the predictive models

Simply run the experiment_runner.py script with the --evaluate flag. If no option is specified the script will train and test automatically the model

Elaborating the results

In order to check improvements between this project and its original implementation, run the parse_results.py script, indicating the folder for the two models before running it.