Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Achieving Group Fairness through Independence in Predictive Process Monitoring

Jari Peeperkorn, Simon De Vos

TL;DR

This work addresses group fairness in Outcome-Oriented Predictive Process Monitoring by enforcing independence between predictions and protected attributes. It introduces threshold-independent distribution-based metrics, ABPC and ABCC, alongside ΔDP_c and ΔDP_b^t, and integrates an Integral Probability Metric loss using the Wasserstein distance into a composite objective \(\mathcal{L}_{\text{total}}=(1-\lambda)\mathcal{L}_{\text{BCE}}+\lambda\mathcal{L}_{\text{IPM}}\) to balance fairness with predictive accuracy. The authors validate the approach through proof-of-concept experiments on artificial event logs (hiring, lending, renting) using LSTM classifiers, demonstrating trade-offs and Pareto fronts between fairness and performance, and highlighting the practicality of threshold-free fairness evaluation in PPM. They provide practical guidelines and discuss limitations, such as reliance on synthetic data and calibration concerns, while outlining future directions including additional fairness definitions, continuous sensitive attributes, and multiclass tasks.

Abstract

Predictive process monitoring focuses on forecasting future states of ongoing process executions, such as predicting the outcome of a particular case. In recent years, the application of machine learning models in this domain has garnered significant scientific attention. When using historical execution data, which may contain biases or exhibit unfair behavior, these biases may be encoded into the trained models. Consequently, when such models are deployed to make decisions or guide interventions for new cases, they risk perpetuating this unwanted behavior. This work addresses group fairness in predictive process monitoring by investigating independence, i.e. ensuring predictions are unaffected by sensitive group membership. We explore independence through metrics for demographic parity such as $Δ$DP, as well as recently introduced, threshold-independent distribution-based alternatives. Additionally, we propose a composite loss function existing of binary cross-entropy and a distribution-based loss (Wasserstein) to train models that balance predictive performance and fairness, and allow for customizable trade-offs. The effectiveness of both the fairness metrics and the composite loss functions is validated through a controlled experimental setup.

Achieving Group Fairness through Independence in Predictive Process Monitoring

TL;DR

This work addresses group fairness in Outcome-Oriented Predictive Process Monitoring by enforcing independence between predictions and protected attributes. It introduces threshold-independent distribution-based metrics, ABPC and ABCC, alongside ΔDP_c and ΔDP_b^t, and integrates an Integral Probability Metric loss using the Wasserstein distance into a composite objective \(\mathcal{L}_{\text{total}}=(1-\lambda)\mathcal{L}_{\text{BCE}}+\lambda\mathcal{L}_{\text{IPM}}\) to balance fairness with predictive accuracy. The authors validate the approach through proof-of-concept experiments on artificial event logs (hiring, lending, renting) using LSTM classifiers, demonstrating trade-offs and Pareto fronts between fairness and performance, and highlighting the practicality of threshold-free fairness evaluation in PPM. They provide practical guidelines and discuss limitations, such as reliance on synthetic data and calibration concerns, while outlining future directions including additional fairness definitions, continuous sensitive attributes, and multiclass tasks.

Abstract

Predictive process monitoring focuses on forecasting future states of ongoing process executions, such as predicting the outcome of a particular case. In recent years, the application of machine learning models in this domain has garnered significant scientific attention. When using historical execution data, which may contain biases or exhibit unfair behavior, these biases may be encoded into the trained models. Consequently, when such models are deployed to make decisions or guide interventions for new cases, they risk perpetuating this unwanted behavior. This work addresses group fairness in predictive process monitoring by investigating independence, i.e. ensuring predictions are unaffected by sensitive group membership. We explore independence through metrics for demographic parity such as DP, as well as recently introduced, threshold-independent distribution-based alternatives. Additionally, we propose a composite loss function existing of binary cross-entropy and a distribution-based loss (Wasserstein) to train models that balance predictive performance and fairness, and allow for customizable trade-offs. The effectiveness of both the fairness metrics and the composite loss functions is validated through a controlled experimental setup.

Paper Structure

This paper contains 19 sections, 5 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Background
  3. Preliminaries
  4. Related Works
  5. Group Fairness through Independence
  6. Demographic Parity Metrics
  7. Integral Probability Metric Loss
  8. Methodology
  9. Preprocessing
  10. Model
  11. Hyperparameters
  12. Experimental Setup & Results
  13. Datasets
  14. Experiment 1: Assessing Fairness Metrics
  15. Experiment 2: Testing IPM Loss
  16. ...and 4 more sections

Figures (5)

  • Figure 1: A toy example showing the need for threshold-free independence metrics.
  • Figure 2: A graphical depiction of the LSTM model used.
  • Figure 3: The LSTM propensity densities for the $hiring$ event log.
  • Figure 4: The results for including IPM loss with different values for $\lambda$.
  • Figure 5: The LSTM propensity densities for the hiring_high event log.