Analyzing and Addressing Data-driven Fairness Issues in Machine Learning Models used for Societal Problems

Aug 4, 2026

This work systematically analyzes and addresses fairness issues in machine learning models caused by class imbalances in data used for societal problem settings. Spectral analysis is employed first to provide evidence and characterize the fairness problems. Class-imbalance correction techniques are then applied prior to training multiple models. Model performance is evaluated using several metrics, followed by comparative analysis of approach merits. Results show oversampling does not universally correct bias; Majority Weighted Minority Oversampling (MWMOTE) performs best on measured fairness and accuracy for the dataset, while some techniques degrade both performance and fairness.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

AccessFree to download

Introduction

Fairness issues from data imbalance

Societal impact and protected attributes

Fairness metrics and performance trade-off

San Jose State University SJSU ScholarWorks Faculty Research, Scholarly, and Creative Activity 1-21-2023 Analyzing and Addressing Data-driven Fairness Issues in Machine Learning Models used for Societal Problems Vishnu S. Pendyala San Jose State University, [vishnu.pendyala@sjsu.edu](vishnu.pendyala@sjsu.edu) HyungKyun Kim San Jose State University Follow this and additional works at: [https://scholarworks.sjsu.edu/faculty_rsca](https://scholarworks.sjsu.edu/faculty_rsca) Recommended Citation Vishnu S. Pendyala and HyungKyun Kim. "Analyzing and Addressing Data-driven Fairness Issues in Machine Learning Models used for Societal Problems" 2023 International Conference on Computer, Electrical & Communication Engineering (ICCECE) (2023) . [https://doi.org/10.1109/](https://doi.org/10.1109/)[ ](https://doi.org/10.1109/)ICCECE51049.2023.10085470 This Conference Proceeding is brought to you for free and open access by SJSU ScholarWorks. It has been accepted for inclusion in Faculty Research, Scholarly, and Creative Activity by an authorized administrator of SJSU ScholarWorks. For more information, please contact [scholarworks@sjsu.edu](scholarworks@sjsu.edu). Analyzing and Addressing Data-driven Fairness Issues in Machine Learning Models used for Societal Problems 1st Vishnu S. Pendyala Department of Applied Data Science San Jose State University San Jose, United States [vishnu.pendyala@sjsu.edu](vishnu.pendyala@sjsu.edu) 2nd HyungKyun Kim Department of Computer Science San Jose State University San Jose, United States [hyungkyun.kim@sjsu.edu](hyungkyun.kim@sjsu.edu) Abstract—This work aims to systematically analyze and address fairness issues arising in machine learning models because of class imbalances present in data, specifically used for addressing societal problems and providing unique insights. Using a specific data set, spectral analysis is first performed to present evidence and characterize the fairness issues. Subsequently, a series of class imbalance correction techniques are applied before the data is used to generate various machine learning models. The models so generated are then evaluated using multiple metrics. The results are then analyzed to compare the various approaches to determine the relative merits of each. As the experiments described in this paper confirm, not all oversampling techniques help in correcting data-induced model biases. Based on the Kappa statistic, F-1 score, and accuracy measured by the area under the Receiver Operating Characteristic curve, among the approaches evaluated, the Majority Weighted Minority Oversampling Technique, MWMOTE oversampling technique addresses the fairness issues the best and also improves the performance of the models at least for the dataset in consideration. The experiments also demonstrate that some of the oversampling techniques can degrade the models both in terms of performance and fairness. The results are interpreted using the evaluation metrics. Index Terms—Machine Learning, Model Fairness, Bias, Fairness Metrics, Class Imbalance, Cohen’s Kappa statistic, Majority Weighted Minority Oversampling Technique I. INTRODUCTION A simple search on Google for images of ”people who are cocaine addicted” unfairly returns images of mostly white people as if it is the white people who are mostly addicted to cocaine. Obviously, the search is biased by the data collected on the internet. Imbalanced data skew the predictions, good or bad, based on the data, in favor of the majority class. The problem is much more poignant in machine learning models, which are increasingly being used for policy-making, regulation, and other wide-impacting applications [1] using legally protected or sensitive attributes such as ethnicity, gender, and religion. The number of AI/ML related bills in the United States has been growing steeply indicating the increasing role of these technologies in our day-to-day lives and well-being. Several instances of ser

FAQs

What fairness issues does the paper focus on in machine learning models?

The paper focuses on data-driven biases caused by class imbalances in datasets used for societal problem applications.

How do the authors detect and characterize fairness problems before modeling?

They perform spectral analysis on a specific dataset to present evidence and characterize the fairness issues.

Which oversampling approach is reported as most effective?

The Majority Weighted Minority Oversampling Technique (MWMOTE) is reported to address fairness issues best while improving model performance for the dataset considered.

Aug 4, 2026

Upload by Emma Wilson