Techniques for Mitigating Bias in Machine Learning Algorithms in the Construction Industry

Machine learning is transforming the construction industry by enhancing efficiency, optimizing decision-making, and improving project outcomes. However, concerns regarding fairness and bias within ML algorithms pose ethical challenges, particularly in South Africa, where historical and socioeconomic disparities could exacerbate inequitable outcomes. This study examines the effectiveness of various techniques in mitigating ML algorithm biases when applied in the construction industry. Using a comprehensive literature review and the Delphi method, insights were gathered from thirteen experts with expertise in construction digitalization and AI adoption. Statistical measures such as mean scores, medians, and interquartile deviations were used to evaluate the most effective bias mitigation techniques. The findings reveal that Model Selection and Evaluation, Hybrid Techniques (combining pre-, in- and post-processing methods), and Fairness-aware Learning Frameworks rank as the most effective approaches. In contrast, Pre-processing and In-processing methods were found to be the least effective. This paper recommends that stakeholders prioritize continuous model evaluation, fairness-aware learning frameworks, and hybrid mitigation strategies to minimize bias in machine learning (ML) applications. It also encourages stakeholder collaboration and the development of regulatory frameworks and industry-wide best practices to promote transparency, accountability, and inclusivity in South Africa’s construction sector.