Decolonising Bias in Organisational Systems: A Machine Learning Approach to Equity, Power, and Algorithmic Justice

This chapter examines how algorithmic systems can reproduce and exacerbate structural inequities along gender and racial lines, using the Adult Income dataset as a testbed for comparative analysis across four models: Logistic Regression, XGBoost, Explainable Boosting Machines (EBM), and Adversarial Debiasing Networks. Empirical evaluation revealed substantial disparities in unmitigated models, with disparate impact ratios falling as low as 0.68 for women and non-white individuals. Crucially, this study embeds technical findings within a decolonial theoretical framework, arguing that fairness cannot be reduced to statistical parity. Instead, it must be understood as a historically situated, epistemically accountable, and relationally constructed concept. The research challenges dominant narratives of algorithmic neutrality by foregrounding the colonial legacies and institutional hierarchies that inform both data practices and model design. By bridging machine learning evaluation with critical social theory, this research advances a reflexive, justice-oriented approach to algorithmic governance in organisations. It offers a framework for rethinking fairness not simply as a computational objective, but as a moral and organisational commitment grounded in equity, participatory design, and the inclusion of marginalised knowledges.