Today, machine learning is integrated into computer engineering systems like cyber security, embedded intelligence, computer vision pipelines, software analytics, networked infrastructures and autonomous decision support environments. In many applications these systems are required to perform in evolving, competitive and impactful environments, and the traditional metrics of accuracy, precision, recall and computational efficiency are no longer adequate for assessing the quality of the model. A reliable machine learning system should also be robust to perturbations, distributional uncertainty, be fair to all affected individuals and groups, and be interpretable for debugging, audit, contestability, and governance. It explores trustworthy machine learning from the three main perspectives of robustness, fairness and interpretability, and places them in the context of the engineering lifecycle of data preparation, model development, validation, deployment, monitoring, and accountability. The article is structured into a synthesis of 47 references published since 2010 and concludes that trustworthiness should be considered more as an engineering requirement at the system level, rather than a posteriori ethical or regulatory corrective. The review also reveals that there is a strong interdependency between robustness, fairness and interpretability. Changes to one dimension can introduce tensions to another, and a holistic (integrated) evaluation is needed. It ends with a proposal for a lifecycle-based agenda for trustworthy machine learning in computer engineering, for which it is crucial to benchmark from multiple perspectives, to validate the system in deployment settings, to be accountable for the system design, and to manage cross-dimensional trade-offs explicitly.
Publication Date: 2026-06-15