Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Fairness-Enhancing Ensemble Classification in Water Distribution Networks

Janine Strotherm, Barbara Hammer

TL;DR

It is demonstrated that typical methods for the detection of leakages in WDNs are unfair in this sense and a remedy to increase the fairness is proposed which can be applied even to non-differentiable ensemble classification methods as used in this context.

Abstract

As relevant examples such as the future criminal detection software [1] show, fairness of AI-based and social domain affecting decision support tools constitutes an important area of research. In this contribution, we investigate the applications of AI to socioeconomically relevant infrastructures such as those of water distribution networks (WDNs), where fairness issues have yet to gain a foothold. To establish the notion of fairness in this domain, we propose an appropriate definition of protected groups and group fairness in WDNs as an extension of existing definitions. We demonstrate that typical methods for the detection of leakages in WDNs are unfair in this sense. Further, we thus propose a remedy to increase the fairness which can be applied even to non-differentiable ensemble classification methods as used in this context.

Fairness-Enhancing Ensemble Classification in Water Distribution Networks

TL;DR

It is demonstrated that typical methods for the detection of leakages in WDNs are unfair in this sense and a remedy to increase the fairness is proposed which can be applied even to non-differentiable ensemble classification methods as used in this context.

Abstract

As relevant examples such as the future criminal detection software [1] show, fairness of AI-based and social domain affecting decision support tools constitutes an important area of research. In this contribution, we investigate the applications of AI to socioeconomically relevant infrastructures such as those of water distribution networks (WDNs), where fairness issues have yet to gain a foothold. To establish the notion of fairness in this domain, we propose an appropriate definition of protected groups and group fairness in WDNs as an extension of existing definitions. We demonstrate that typical methods for the detection of leakages in WDNs are unfair in this sense. Further, we thus propose a remedy to increase the fairness which can be applied even to non-differentiable ensemble classification methods as used in this context.

Paper Structure

This paper contains 19 sections, 2 theorems, 14 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Fairness in Machine Learning
  3. Leakage Detection in Water Distribution Networks
  4. Methodology
  5. Virtual Sensors
  6. Ensemble Leakage Detection
  7. Evaluation
  8. Application Domain and Data Set
  9. Experimental Results and Analysis: Residual-Based Ensemble Leakage Detection Does Not Obey Fairness
  10. Fairness-Enhancing Leakage Detection in Water Distribution Networks
  11. Methodology
  12. Optimizing Loss with Fairness Constraints
  13. Optimizing Fairness with Accuracy Constraints
  14. Evaluation
  15. Experimental Results and Analysis
  16. ...and 4 more sections

Key Result

Lemma 2.1

Let $S_k$ be the sensitive feature describing whether a leakage is active in the protected group $k$ of the for each $k = 1,...,K$. Moreover, let $\epsilon, \tilde{\epsilon} \in [0,1]$ and define $\max_k := \max_{ k \in \{1,...,K\} } \mathbb{P}(\hat{Y}=1 ~|~ S_{k}=1)$.

Figures (4)

  • Figure 2.1: The Hanoi , its sensor nodes (IDs 3, 10 and 25) and the protected groups, each highlighted in another color (group 1 on the left side in light shade, group 2 in the middle in dark shade, group 3 on the right side in middle shade). The sensor nodes are colored in the same color of the protected group they belong to and highlighted with a grey circle.
  • Figure 3.1: Accuracy and disparate impact score per method and leakage diameter.
  • Figure 3.2: Coherence of accuracy and disparate impact score for the different fairness-enhancing methods and different leakage sizes. The cross data points visualize the disparate impact score and accuracy of the non-fairness-enhancing baselines (T-F-PR, dark blue, for T-F-PR+F and ACC, light blue, for ACC+F and DI+ACC).
  • Figure 3.3: Coherence of accuracy, disparate impact and the training hyperparameter.

Theorems & Definitions (6)

  • Remark 1.1
  • Lemma 2.1: Equivalence of disparate impact and equal opportunity in
  • proof
  • Corollary 2.2
  • proof
  • Remark 3.1