Sustainable Selection of Machine Learning Algorithm for Gender-Bias Attenuated Prediction

Research into novel approaches like Machine Learning (ML) promotes a new set of opportunities for sustainable development of applications through automation. However, there are certain ML tasks which are prone to spurious classification, mainly due to the bias in legacy data. One well-known and high...

Full description

Saved in:
Bibliographic Details
Main Authors: Raik Orbay, Evelina Wikner
Format: Article
Language:English
Published: IEEE 2025-01-01
Series:IEEE Open Journal of Vehicular Technology
Subjects:
Online Access:https://ieeexplore.ieee.org/document/10759097/
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1832590364571074560
author Raik Orbay
Evelina Wikner
author_facet Raik Orbay
Evelina Wikner
author_sort Raik Orbay
collection DOAJ
description Research into novel approaches like Machine Learning (ML) promotes a new set of opportunities for sustainable development of applications through automation. However, there are certain ML tasks which are prone to spurious classification, mainly due to the bias in legacy data. One well-known and highly actual misclassification case concerns gender. As the vast dataset for engineering rules, standards and experiments are based on men, a bias towards women is the subject of research. Accordingly, any bias should be contained before the algorithms are deployed to the service of the sustainable society. There is a substantial amount of data on ML gender-bias in the literature. In these, the majority of the investigated cases are for ML branches like image or sound processing and text recognition. However, utilizing ML for driving style investigations is not an extensively researched area. In this work, a novel application for gender-based classification with bias-attenuation using anonymized driving data will be presented. Using data devoid of biometric and geographic information, the proposed pipeline distinguishes manifested binary genders with 80% accuracy for the drivers in the holdout data set. In addition, a method for sustainable algorithm selection and its extension to embedded applications, is proposed. An investigation into the environmental burden of seven different types of ML algorithms was conducted and the popular neural network algorithm had the highest environmental burden.
format Article
id doaj-art-e30186a919ce468d915d8d56021b0600
institution Kabale University
issn 2644-1330
language English
publishDate 2025-01-01
publisher IEEE
record_format Article
series IEEE Open Journal of Vehicular Technology
spelling doaj-art-e30186a919ce468d915d8d56021b06002025-01-24T00:02:19ZengIEEEIEEE Open Journal of Vehicular Technology2644-13302025-01-01634835810.1109/OJVT.2024.350292110759097Sustainable Selection of Machine Learning Algorithm for Gender-Bias Attenuated PredictionRaik Orbay0https://orcid.org/0000-0002-0339-1807Evelina Wikner1https://orcid.org/0000-0002-7203-6243Volvo Car Corporation - 97100 Propulsion and Energy - Strategy & Execution, Gothenburg, SwedenDepartment of Electrical Engineering, Chalmers Technology University, Gothenburg, SwedenResearch into novel approaches like Machine Learning (ML) promotes a new set of opportunities for sustainable development of applications through automation. However, there are certain ML tasks which are prone to spurious classification, mainly due to the bias in legacy data. One well-known and highly actual misclassification case concerns gender. As the vast dataset for engineering rules, standards and experiments are based on men, a bias towards women is the subject of research. Accordingly, any bias should be contained before the algorithms are deployed to the service of the sustainable society. There is a substantial amount of data on ML gender-bias in the literature. In these, the majority of the investigated cases are for ML branches like image or sound processing and text recognition. However, utilizing ML for driving style investigations is not an extensively researched area. In this work, a novel application for gender-based classification with bias-attenuation using anonymized driving data will be presented. Using data devoid of biometric and geographic information, the proposed pipeline distinguishes manifested binary genders with 80% accuracy for the drivers in the holdout data set. In addition, a method for sustainable algorithm selection and its extension to embedded applications, is proposed. An investigation into the environmental burden of seven different types of ML algorithms was conducted and the popular neural network algorithm had the highest environmental burden.https://ieeexplore.ieee.org/document/10759097/Battery electric vehiclesdriving style classificationenergy consumptionfeature engineeringmachine learningunsupervised/supervised learning
spellingShingle Raik Orbay
Evelina Wikner
Sustainable Selection of Machine Learning Algorithm for Gender-Bias Attenuated Prediction
IEEE Open Journal of Vehicular Technology
Battery electric vehicles
driving style classification
energy consumption
feature engineering
machine learning
unsupervised/supervised learning
title Sustainable Selection of Machine Learning Algorithm for Gender-Bias Attenuated Prediction
title_full Sustainable Selection of Machine Learning Algorithm for Gender-Bias Attenuated Prediction
title_fullStr Sustainable Selection of Machine Learning Algorithm for Gender-Bias Attenuated Prediction
title_full_unstemmed Sustainable Selection of Machine Learning Algorithm for Gender-Bias Attenuated Prediction
title_short Sustainable Selection of Machine Learning Algorithm for Gender-Bias Attenuated Prediction
title_sort sustainable selection of machine learning algorithm for gender bias attenuated prediction
topic Battery electric vehicles
driving style classification
energy consumption
feature engineering
machine learning
unsupervised/supervised learning
url https://ieeexplore.ieee.org/document/10759097/
work_keys_str_mv AT raikorbay sustainableselectionofmachinelearningalgorithmforgenderbiasattenuatedprediction
AT evelinawikner sustainableselectionofmachinelearningalgorithmforgenderbiasattenuatedprediction